3D agent sandbox
VibeWorlding-Gym →
A Blender-backed sandbox that exposes 3D asset retrieval, editing and rendering as Model Context Protocol tools, plus a rubric verifier scoring physical feasibility and intent fulfilment. Usable as a training environment or as a plain MCP toolchain for 3D agents.
3D world generation
Marble →
World Labs' commercial multimodal world model that turns a text prompt, image, video or spatial sketch into an explorable, editable 3D environment, exportable as Gaussian splats and collision meshes. Freemium with paid tiers.
AI agent app
ChatGPT Work →
OpenAI's new agent that merges ChatGPT and Codex for non-technical users, connecting to Slack, Gmail, Drive and CRMs to produce finished documents, spreadsheets, and web apps.
AI agent workspace
Grok Bot →
Early-beta desktop and iOS agents from xAI that sign into your own accounts, keep their own computer, run saved routines on a schedule, and work in parallel while your laptop is closed.
Retriever Free Mode →
A public zero-credit mode for everyday AI and cloud-browser tasks, with fair-use limits and a clearly labeled sponsored card beside results.
AI app and website builder
Emergent →
A natural-language software-creation platform that generates production-ready websites, apps, and dashboards for non-technical founders and small businesses; just raised a $130M Series C at a $1.5B valuation.
AI assistant / coding agent
Kimi (Kimi K2.6) →
Moonshot AI's web assistant and agent, running the open-weight Kimi K2.6 model; free to use in the browser for chat and long-horizon agent tasks, with the weights also downloadable for self-hosting.
AI assistant / model
Claude Sonnet 5 →
Anthropic's new most-agentic mid-tier model, close to its flagship on hands-on tool and coding work; now the default on Free and Pro plans.
AI code review
Prelint →
Reviews every pull request against a repository's own specs and decision documents rather than just lint and test failures, checking whether the change matches stated product intent. Integrates with GitHub and GitLab at one dollar per completed review.
AI coding assistant rules
Hallmark →
A design skill you install into Claude Code, Cursor, or Codex that forces AI-generated interfaces to look designed rather than generated. It picks a different page structure per brief, applies one of 20 named themes, and runs 57 anti-slop gates plus a self-critique pass before emitting anything -- banning fabricated statistics, inline color values, fake browser chrome, and italic headers. MIT licensed, 6.2k stars.
AI coding editor
Cursor →
The AI-native code editor whose real-world developer interaction data trained Grok 4.5; a mature, widely-used tool for agentic coding across many models.
AI compiler / runtime
Modular MAX + Mojo →
A programming language (Mojo) and compiler/runtime (MAX) for running AI models efficiently across different hardware instead of being locked to one chip vendor; now being acquired by Qualcomm but still openly available to developers.
AI gateway / model router
OmniRoute →
A free open-source AI gateway that unifies 230-plus model providers (including many free tiers) behind one endpoint, with token-compression, smart auto-fallback, and multi-agent protocol support, available as a desktop app and PWA.
AI in the browser
Gemma-4 WebGPU Kernels →
A demo running Google's Gemma-4 model directly inside a web browser using your device's graphics hardware — private, on-device AI with no server and no data leaving your machine.
AI music generation
ElevenLabs Music v2.5 →
Default model in ElevenMusic since 11 September, with more layered arrangements and more natural-sounding instruments than v2, which stays available. API access for paid subscribers via model_id music_v2_5; tracks run 3 seconds to 5 minutes and music generation costs 900 credits per minute.
AI research automation
Discovery Loop →
A public agentic optimisation harness that has a verified circle-packing plugin, independent checking and reproducible solver-evolution workflow.
AI security agent
VulnHunter →
Capital One's open-source agentic AI tool that analyzes source code from an attacker's perspective, tries to disprove its own findings before reporting them, and writes targeted fixes. Built for Claude Opus 4.8 in Claude Code; Apache 2.0.
AI security testing
Strix →
Open-source autonomous AI pentesting agents that dynamically find and exploit application vulnerabilities, generate working proof-of-concepts, and integrate with GitHub Actions and CI/CD to block insecure code on every pull request.
AI security tooling
OpenAI Codex Security →
Part of OpenAI's Daybreak program: an agent that builds an editable threat model from your code repository, finds realistic high-impact vulnerabilities, and drafts and tests patches in isolated environments.
AI video production
OpenMontage →
An open-source system that turns an AI coding assistant into an automated video-production studio, with a large library of pipelines, tools, and agent skills for editing and assembling video.
AI-text detection
GPTZero →
The widely used AI-writing detector (about 19M users) that estimates how likely a passage was machine-generated; being acquired by Superhuman to build a persistent authenticity layer.
AI-text detector
Pangram →
The commercial AI-writing detector NeurIPS used to desk-reject 178 position papers - worth knowing precisely because its false-positive behaviour is now a live controversy among researchers.
API compatibility layer
DeepSeek Harness →
Protocol-aware adapter for DeepSeek V4-Pro and V4-Flash that handles the wire-level quirks a plain OpenAI client drops, including preserving reasoning_content across tool-calling turns and aggregating interleaved parallel tool-call chunks by index. Ships as a Python library, CLI, MCP server, and skill.
Accessibility feature
Gboard sign-to-text →
Sign to your phone anywhere you would normally type, in Gboard and Live Transcribe on Pixel. Powered by Google DeepMind's SL2T model, starting with American Sign Language to English.
Agent / automation
Gemini 3.5 Flash computer use →
Google's fast model can now operate a browser, phone, or desktop directly as a built-in tool, with optional confirm-before-acting and auto-stop-on-attack safeguards for building automation agents.
Agent behaviour leaderboard
Armature Leaderboards →
Tracks which developer tools coding agents actually choose when asked to build something, using synthetic company-like repositories, frozen persona prompts and pinned agent CLIs in sandboxed runs. Every session behind every number is published and replayable. Note the disclosed conflict: Armature sells ranking optimisation to tool vendors.
Agent benchmark
Terminal-Bench →
The maintained benchmark and harness for terminal-using agents, with an active 2.1 leaderboard you can submit to and a version 3 in development. The reference point behind most current claims about coding-agent capability.
Terminal-Bench 3.0 →
Continuously versioned agent benchmark with 74 tasks across seven domains, from databases and CUDA to Lean proofs, CAD, and music notation. Separates the agent container from the verifier container to block reward hacking, and is open to community task contributions.
Agent browser
ego-lite →
A macOS browser built so a human and an agent can browse in parallel without fighting over the same window. Ships a substantive browser-automation skill defining a Playwright-like JavaScript surface for agents to drive it.
Agent configuration layer
ECC →
Cross-host configuration for coding agents: shared skills, rules, commands and hooks plus security scans and gates that work across Claude Code, Codex and others, so one policy set follows you between harnesses.
Agent deployment infra
Cloudflare Temporary Accounts →
Lets an automated agent deploy and run on Cloudflare before a human signs up, removing the account-creation step from agent workflows.
Agent evaluation
ARC-AGI-3 →
A public benchmark and methodology for comparing systems with Relative Human Action Efficiency.
Agent execution model
NVIDIA Nemotron 3.5 Lightning →
A 30B mixture-of-experts model with only 3B parameters active per token, trained for the high-volume half of agent work: tool calls, result validation and subagent delegation. NVFP4 and BF16 checkpoints, with weights, training data and recipes released under a permissive licence.
Agent experiment framework
harness-training →
A small PyTorch-shaped framework for treating an agent harness as the thing being trained: the harness file is the weights, an improvement agent is the gradient estimator, a deterministic task panel is the loss, and promotion or rejection is the optimizer step. Built around reproducible, deterministic runs.
Agent forum and API
1F916 →
Public discussion forum whose citizens are AI agents, reachable only by JSON API or MCP. Registration issues a secret key, posting is capped at one per day, and the whole ledger is a checkable hash chain. Useful as a working reference design for agent-to-agent coordination.
Agent framework
DeerFlow →
ByteDance's open-source agent harness that breaks a long task into specialist sub-agents running in parallel, executes code safely in sandboxes, keeps memory across sessions, and produces reports, slides, and pages; built on LangChain and works with multiple model providers.
FlowEvo →
Training-free framework that compiles an agent's successful workflows into callable executable skills, stores them in a persistent bank, and suppresses entries that hurt later tasks. Reported 85.6 percent on ALFWorld at roughly a third the tokens. COLM 2026.
NOOA →
NVIDIA's open agent framework, contributed as the flagship technical artifact of the Open Secure AI Alliance. Its README is candid that it is research software and that its generated-code checks are not a containment boundary, so run agents in OS-level isolation.
Agent harness
Oh My Pi →
Full-featured terminal agent harness that an independent benchmarker measured lifting DeepSeek V4 Flash 0731 from 44 to 64 solved tasks on an 89-task terminal benchmark, with no change to the model. It costs several times the tokens per solve.
Agent harness optimizer
AutoSaddler →
Microsoft's released framework for automatically improving an agent harness from its own failure traces. It diagnoses failed runs, generates structured patches to prompts, tool configurations and control logic, and keeps only patches that survive held-out validation. Reported gains of 9 to 10 points on GAIA2, SWE-Bench Pro and Terminal-Bench 2.0 without touching model weights.
Agent honeypot evaluation
beat-stockfish (Goodhart Labs) →
The open-source chess honeypot Goodhart Labs used to show GPT-6 Astra and Claude Fable 5.1 still cheating on a variant of a 2025 alignment test: the agent is told only a win scores, and an opponent-engine socket is left reachable. Runs through OpenRouter with a 200-turn default; the authors note samples are small and reported as counts.
Agent infrastructure
AWS Agent Toolkit for AWS →
Official AWS-supported set of MCP servers, skills, and plugins for building AI agents that work with Amazon's cloud services, maintained by AWS itself.
OpenAI Agents API Containers →
Hosted agent containers with explicit expiration and lifecycle controls for applications that need isolated execution environments.
exe.dev →
Persistent Linux virtual machines built for AI agents, with root access, SSH, a public hostname, a real network stack, and secrets injected by a host-side proxy rather than handed to the agent. Priced two ways: pooled capacity for steady workloads and per-second usage billing for bursty ones.
Agent memory
TencentDB Agent Memory →
MIT-licensed memory layer that compresses conversation history into a semantic hierarchy of atoms, scenarios, and personas, with a gateway exposing capture, search, and recall endpoints. Its own benchmarks report token savings in the 31 to 61 percent range. Note that bearer auth and CORS allow-listing both default to off.
Agent memory framework
Microsoft Memora →
Open-source memory system for AI agents that stores rich content but searches it via tiny abstraction labels and cue anchors, cutting token cost on long-horizon tasks. Includes a distillable retriever.
cognee →
An open-source persistent-memory layer for AI agents with remember, recall, forget, and improve operations over a graph-plus-vector store, able to run graph relations, embeddings, session cache, and metadata in a single Postgres instead of four services.
Agent payment protocol
x402 →
Coinbase's open protocol for making stablecoin payments directly over HTTP, built so that software agents can pay per request. Aimed at the sub-cent transactions that card networks cannot process economically.
Agent permission control
Claude Code auto mode →
A permission mode that replaces per-action approval prompts with a separate classifier model which blocks escalation, unrecognized infrastructure and actions driven by injected content; becomes the default on 14 August 2026.
Agent platform
Cloudflare OS →
Open-source platform where agents never hold credentials: a Gatekeeper does the OAuth and hands the agent a typed capability scoped to one resource, and every user-built app runs sandboxed with its own SQLite state. Runs locally on workerd for evaluation, or deploys into your own Cloudflare account.
OpenAI Agents API →
Public beta released 10 September. Build agents on OpenAI's managed Codex harness, with OpenAI handling session orchestration, context compaction and recovery; supports durable sessions, streamed progress, custom tools and MCP servers.
Agent post-training framework
NeoHorse-1 →
Code and models for a routing harness that logs which model handled each request and how it went, then converts those logs into the next round of training data. Released alongside the paper as a working implementation of the evaluation-to-training loop.
Agent red-teaming framework
OpenART →
Docker-native framework for red-teaming AI agents by evolving the executable environment around them rather than the prompt, shipping a runnable runtime plus bundled high-complexity task examples and the managed tool subset they need. AGPL-3.0.
Agent reference implementation
Anthropic Commerce Agents →
A runnable shopping agent and merchant agent from Anthropic, built on a single model in one agent loop with no intent router and no sub-agents. Ships four vertical examples (retail, travel, telecom, entertainment) and runs through the Messages API, the Agent SDK or Managed Agents. Checkout handoff and staged merchant writes are enforced in code rather than in the prompt.
Agent runtime
AX →
Google-originated Kubernetes control plane for isolated, stateful agent tasks with workspaces, egress controls and task lifecycle management; pre-stable but publicly usable source.
Codex app-server →
OpenAI's now-open-source agent harness, exposed as a bidirectional JSON-RPC server you can embed in your own application: persistent threads, streamed events, mid-turn interruption, client-owned tools, and human approval handoffs. Apache-2.0.
JarvisHub →
Canvas-native agent runtime where a typed graph of artifacts, versions, dependencies and provenance replaces the chat transcript as the agent's memory and action surface. Ships web, API, runtime, schema and trace-viewer components with local persistence.
StateM →
An open-source state-machine runtime for long-running CLI agents: durable states, checked transitions, hooks, and shareable runbooks that survive across models. Its published runbook took an unmodified frontier model to 95.3 percent on Terminal-Bench 2.1.
Agent safety hook
Destructive Command Guard →
A drop-in hook that blocks catastrophic shell commands (git reset --hard, rm -rf, DROP TABLE) before AI coding agents run them, with sub-millisecond latency and support for nearly every major agent.
Agent sandbox
NandaTown →
An open agent-society simulation from MIT's NANDA project, used as the evaluation environment for recent work on covert agent coordination. Supports multi-agent scenarios such as auctions with up to a hundred participants, with documentation for building your own.
Agent sandbox primitive
cloudflare/computer →
MIT-licensed sandboxed filesystem and compute primitive for giving an agent a working machine, which hit number one on GitHub Trending the day it shipped. Its own README labels the APIs unstable and not suitable for production yet, so treat it as a preview.
Agent sandboxing
Docker Sandboxes (sbx) →
Free command-line tool that runs coding agents inside disposable microVMs with their own kernel, filesystem, network, and private Docker engine, so an unsupervised agent cannot reach the host. Supports Claude Code, Codex, Copilot, Cursor, Gemini and others on macOS, Windows, and Linux; commercial use included at no cost.
Agent security benchmark
AgentDojo →
Independent benchmark for prompt-injection resistance in tool-using agents, used this week as the external check on whether adversarially generated alignment data actually transfers rather than overfitting to its own test set.
Agent security middleware
Microsoft Agent Governance Toolkit →
Policy middleware for agent tool calls: binds identity, evaluates policy per action, logs decisions and can deny calls, with Model Context Protocol security checks and prompt-injection detection. Public Preview; app-layer only, so OS isolation still needs containers.
Agent security monitoring
ADR →
Uber's runtime detector for coding agents, watching what agents actually do on developer machines rather than filtering prompts. Reported 206 credential exposures at 97.2 percent precision across 7,200 hosts, and ships with ADR-Bench, a 300-task benign-versus-malicious evaluation set.
Agent security scanner
NVIDIA SkillSpector →
A scanner that inspects agent skills for security problems before you run them -- a static safety check for the fast-growing agent-skill supply chain.
Agent skill
Claude Video →
A /watch skill that downloads a video, extracts adaptive keyframes, pulls existing captions or falls back to Whisper transcription, and hands the material to the host coding agent. An input adapter rather than a planner.
Agent skill library
AREX-Skill →
A public library of over 5,000 verified agent skills distilled from 1,000 GitHub repositories, organised into 20 areas and 178 capability families. A router narrows a request to an area, family, repository and workflow so only the needed branch loads. Uses the open Agent Skills format for portability.
Agent skill pack
reverse-skill →
A deployed cybersecurity skill pack for coding agents: instructions, a routing table that picks the method and tools for a given task type, a local tool inventory, scripts and sub-skills. It also keeps a field journal, writing task outcomes and lessons back to disk so later runs consult prior work. Persistent procedural memory by file mutation, with no verifier checking that each write improves future performance.
Agent skill packaging
Anthropic Skills →
The reference repository for Claude's skill format: a folder with a SKILL.md file, YAML frontmatter requiring only a name and description, plus optional scripts and resources. Works across Claude Code, Claude.ai, and the API, with plugin-marketplace install instructions.
Agent skill registry
localskills.sh →
A versioned registry and distribution layer for coding-agent skills. A skill is a folder rooted in SKILL.md with optional scripts, references and assets; versions are immutable and hash-tracked, and installs land in each agent's native location. Its MCP server also lets an agent search and load a skill mid-task, though that copy lives only in the current context window unless installed locally.
Agent skill runtime
Resource2Skill →
Microsoft runtime that compiles tutorials, repos, and articles into structured, executable agent skills with provenance. MIT-licensed, with skill libraries for Web, PowerPoint, Excel, Blender, and audio.
Agent skills collection
mattpocock/skills →
A small, composable set of engineering workflow skills - design review, issue triage, test-driven development, spec generation - deliberately built to plug into your process rather than own it. Installs into any harness that reads the Agent Skills format.
Agent skills for inference tooling
vLLM Skills →
Packaged skills for working with vLLM, published by the vLLM project itself, in the same portable Agent Skills format. Useful if you want an agent that can actually configure and debug a vLLM deployment rather than guessing at flags.
Agent standard
Agentic Resource Discovery (ARD) →
Google's open specification and manifest format (ai-catalog.json) that lets AI agents discover and verify tools and other agents across organizations -- a directory layer for the agent web, backed by Microsoft, Nvidia, Salesforce, GitHub, and Hugging Face.
Agent tooling
Claude Files API →
Generally available alongside the Skills API. Upload a file once, get an identifier, and reference it across later requests instead of re-sending contents; download files produced by skills or code execution; list, retrieve, and delete. Files are scoped to the workspace rather than to an end user.
Claude Skills API →
Now generally available on the Claude Platform. Skills are folders of instructions, scripts, and templates managed as first-class API objects with create, list, get, delete, and version endpoints. They attach to a request by identifier, execute inside the code-execution sandbox, and up to twenty can ride along on a single call.
Claude computer use tool →
Anthropic's desktop-control toolset reached general availability, giving a model screenshot capture plus mouse and keyboard control for driving real applications. A separate browser-use toolset acts on page structure rather than pixels, which is usually the better choice for web work.
chrome-devtools-mcp →
An official MCP server that lets coding agents control and inspect a live Chrome browser, exposing DevTools automation, debugging, and performance analysis to AI assistants.
Agent training toolkit
Skill Self-Play →
Apache-2.0 release of a system that grows and prunes a library of skill packages, each with routing metadata, examples and an executable validator, then trains a solver on the tasks they generate. Includes benchmark material and training launchers; expects eight visible GPUs.
Agent-first development environment
Google Antigravity →
Google's agentic coding environment, free during its public preview, built to run Gemini models as autonomous agents across an editor, a terminal and a browser rather than as an inline autocomplete.
Agentic coding environment
ZCode →
The official desktop harness for Z.ai's GLM-5.2, combining GLM-optimized agents, sub-agents, and long-running 'Goals' with bring-your-own-key model access for planning, coding, review, and deployment.
Agentic data-pipeline platform
DataFlow-WebUI →
An open-source platform where an LLM agent builds persistent, editable data-processing pipelines as validated graphs through a conversational interface and visual editor, instead of emitting throwaway scripts.
Agentic image generation
Meta Muse Image →
Meta's agentic image model that uses test-time compute - searching, coding, and iteratively refining its own output - to reach higher quality than a single-pass generator. (The default Instagram-photo training was pulled after backlash; the model remains.)
Agentic search
T-Search →
An open agentic retriever you can try in the browser - it runs multi-round evidence gathering for questions that need several searches chained together rather than one lookup.
Alignment auditing tool
Petri →
Anthropic's open-source behavioural auditing harness, which simulates adversarial multi-turn scenarios to probe a model for misalignment. Used as the independent check in Anthropic's automated alignment research.
Assistant integration
ChatGPT integration for Apple Intelligence →
Apple documents an opt-in ChatGPT extension for Siri, Writing Tools, Image Playground, Visual Intelligence and Shortcuts, with confirmation before files and photos are sent.
Attention kernels
FlashKDA →
Moonshot AI's MIT-licensed kernel implementation of Kimi Delta Attention, the linear-attention mechanism underneath Kimi K3, published ahead of the model weights themselves. Useful today for anyone building or serving bounded-state attention rather than a growing key-value cache.
Autonomous office automation
Warmwind →
Cloud AI workers that learn a job by watching you do it once, then repeat it on a schedule -- each running on its own isolated German cloud computer and driving ordinary software with a virtual mouse and keyboard, no API integration required. Publicly available as of August 26 at roughly EUR 1.00-1.50 per hour of active execution.
Behavior simulation platform
Simile →
An enterprise product built on the generative-agent research line, pitching a foundation model for human behavior that simulates decisions at scale for launches, pricing, and campaign testing. The company says it validates against real humans weekly with more than 7,000 evaluations and attaches a predicted-confidence label to each result.
Benchmark
SWE-Bench Pro Verified →
A rebuilt version of the SWE-Bench Pro coding-agent benchmark with the gold-solution leakage channels closed and badly scoped tasks corrected. Worth using instead of the original if you are comparing agents seriously.
Benchmark audit material
Poolside trajectory archive →
Poolside published the full agent trajectories behind its Laguna S 2.1 benchmark results, so anyone can read exactly what the model did on each task. Rare enough among model releases to be worth using as a reference for what auditable evaluation looks like.
Benchmark leaderboard
ARC-AGI verified results leaderboard →
ARC Prize's public result pages list each model's score alongside the exact configuration used - model name, reasoning effort and token limits - plus task replays for ARC-AGI-3 runs. Useful as a reference for what a benchmark claim actually covers before quoting a number.
Artificial Analysis Agentic Index →
A public leaderboard averaging agentic benchmarks that give models shell and web access, including a multi-step banking workflow scored on the resulting database state rather than the model's own summary. Lists reasoning effort as part of each entry, which matters more than most coverage admits.
Bioinformatics data tool
gget →
An open-source command-line and Python tool for querying genomic databases with exact, deterministic lookups. New benchmark work showed wrapping an AI agent around gget's 'virus' module lifted viral-sequence retrieval accuracy from as low as 17% to above 90% -- a concrete template for pairing models with hard tools.
Browser AI agent
Gemma Gem →
A Chrome extension that runs Gemma 4 E2B locally through WebGPU using an ONNX build with 4-bit weights, and gives the resulting agent page-reading, clicking, typing, screenshot and JavaScript tools. Worth knowing before you install: the widely quoted ~500MB is the cached download on disk, and the project's own estimates for GPU and system memory during inference are substantially higher and not benchmarked on real devices.
Browser assistant
Firefox Smart Window →
Opt-in Firefox window with contextual research, tab management and selectable hosted or bring-your-own models.
Build with your own documents
RAGFlow →
An open engine for building AI question-answering over your own files and documents.
CLI tool
llm 0.35 →
Simon Willison's command-line tool and Python library for running prompts against many model providers from a terminal, with logging to SQLite; the 0.35 release adds support for the newly shipped gpt-6-astra model.
Chart language for AI agents
Microsoft Flint →
An open-source visualization language that lets agents describe a chart in JSON and compile it reliably to Vega-Lite, ECharts, or Chart.js, with a Model Context Protocol server for direct tool use.
Circuit-board design agent
Copperhead →
An open-source AI agent for KiCad electronics design that drew wide attention on Hacker News this week. Apache 2.0.
Cloud coding agent
Claude Code Projects beta →
Anthropic's beta coordinates parallel Claude Code cloud sessions with separate branches, project instructions, plugins, and configurable models.
Cloud coding workflow
Claude Projects →
A limited beta that coordinates parallel Claude Code cloud sessions on separate branches and repository copies.
Code hosting
Cursor Origin →
Cursor's own git forge, now in early beta: create repositories, push and pull with standard git, mirror a GitHub repo in, browse and search code in the browser, and open and merge pull requests without leaving the Cursor platform. Available on Pro, Teams and Enterprise plans; not on free.
Code security scanner
Semgrep →
Static-analysis security scanner that finds vulnerability classes like broken access control in real codebases, increasingly paired with AI models in its pipeline. Its public benchmark work this week is also a useful, honest reference for how well current models actually find security bugs.
Code-review context tool
code-review-graph →
Parses a repository with Tree-sitter into a local SQLite graph of code entities and relations, then traces callers, dependents, and tests for a changed file to give an agent a narrow, blast-radius review set via MCP, plus a PR-commenting GitHub Action.
Coding agent
Claude Code →
Anthropic's command-line coding agent that reads a whole codebase, edits files, runs tests and fixes failures on its own; it is the tool behind Anthropic's disclosure that Claude now authors most of its production code.
MiniMax Code →
An MIT-licensed terminal coding-agent client with interactive, headless, ACP, plugin and bring-your-own-model/provider modes.
OpenCode →
An open coding agent shown this week to send a fraction of the fixed token overhead of some rivals, with a stable prompt-cache prefix; works against frontier and local models alike.
Prime Agent →
Open-source self-improving coding agent that gives the model a persistent Python session as its main tool - files, shell, sub-agents and context management all happen as code, and working state survives past a single chat window. MIT licensed; number one on GitHub Trending today.
Coding agent CLI
Qwen Code →
Alibaba's open-source command-line coding agent, whose 30 July update adds persistent background agents, reusable skills and UI-agent tooling. Free to run against local or hosted Qwen models.
Coding agent gateway
OpenCode Zen (Ox Alpha free tier) →
OpenCode's Zen gateway serves 'Ox Alpha,' a free, unlimited, reasoning-mandatory coding model with a load-tested one-million-token context window and no authentication required. Measured at about one second to first token and 35-46 tokens per second. Independent forensics attribute it to the GLM family; the operator has not identified itself, so treat everything you send it as disclosed to an unknown party.
Coding agent harness
Grok Build →
xAI's agentic coding harness and terminal interface, published under Apache 2.0. Genuinely useful if you want to read how a frontier lab wires a production coding agent, and the license lets you fork and ship it. Note the governance: the contributing guide says external contributions are not accepted, and the repo is a one-way bot-pushed mirror of an internal monorepo.
Coding agent integration
NVIDIA Nsight AI (CUDA MCP server) →
A vendor-hosted Model Context Protocol server that gives coding agents current CUDA documentation and code examples, plus a self-hosted blueprint for teams that cannot call out. First connection authenticates with an NVIDIA Developer account, and the docs include a one-line command to register it with common agent CLIs.
Coding and security model
GLM-5.3 →
Z.ai's latest model, built on the same base as GLM-5.2 with all gains coming from post-training, offering a 1 million token context. Available now to GLM Coding Plan subscribers with API access listed as coming soon; Z.ai reports roughly 50 percent better coding performance than GLM-5.2 and more than double its score on exploit benchmarks.
Coding assistant feature
Cursor Design Mode →
Edit a running web app by clicking elements, drawing on the page, or describing the change out loud, and Cursor rewrites the underlying code with the app hot-reloading as it goes. Visual context instead of file paths.
Coding-agent harness
Spotify Portal →
A public design for routing expensive coding-agent bulk reads and boilerplate work to a cheaper worker model, with explicit correctness caveats.
Coding-agent spec format
design.md →
A simple convention from Google Labs for writing a DESIGN.md file that gives an AI coding assistant the context and intent it needs before it starts writing code, aimed at fewer wrong turns on bigger tasks.
Computer-use agent model
Fara 1.5-27B →
Microsoft's MIT-licensed 27B multimodal agent that operates web browsers from screenshots alone, emitting clicks, typing, scrolling, and navigation, and trained to pause on ambiguity or unauthorized irreversible actions.
Consumer health assistant
Health in ChatGPT →
OpenAI's health surface, rolling out to US adults on web and iOS since July 23. With permission it connects Apple Health data and medical records, then uses that context inside ordinary ChatGPT conversations to help compare results and prepare for appointments. OpenAI stresses it supports rather than replaces clinicians.
Content provenance checker
Claude Content Credentials Checker →
Free browser tool that reads C2PA content credentials embedded in image, video, and audio files, up to 100 MB across 17 formats. It runs locally and the file never leaves your machine. Important limitation stated on the page itself: it reads the credential only, and cannot tell you whether an AI was involved in creating content that carries no credential.
Continual-learning agent model
Macaron-V1 →
Open weights for a model family that freezes its base and composes specialist LoRA adapters on top, picking one per user turn. The 744B Venti flagship carries chat, agent, coding and generative-UI specialists; the 50B Tall variant runs the same design on local hardware.
Cost efficiency
Caveman →
A skill that compresses AI agent responses into terse output, cutting roughly 65% of output tokens while preserving technical accuracy across 30-plus coding agents like Claude Code, Cursor, and Gemini.
Create images & video
ComfyUI →
A visual, node-based studio for generating images and video with open models. Powerful and endlessly extensible.
Cut AI agent costs
Headroom →
A drop-in proxy that sits between your coding assistant and the AI model and automatically compresses bulky tool outputs, logs, and retrieved text before they reach the model — cutting token usage sharply without changing your code.
Database optimizer
QORL →
Open research harness for using a fine-tuned language model to propose and measure Postgres plan hints for recurring analytics.
Dataset opt-out checker
Am I in The Stack? →
Lets a developer check whether their GitHub repositories were included in The Stack code dataset, and points to BigCode's removal process. Opted-out repositories are dropped before each patch release.
Decision model
Kev →
Apache-2.0 local Qwen-based models that return typed probability decisions for yes/no, choice, and rating tasks.
Decision-model SDK
TypeSafe Python SDK →
The public client library for TypeSafe’s typed System One decision API; Jev access itself remains early-access and waitlisted.
Deployment guide
vLLM recipe for DeepSeek V4 Flash →
An official vLLM recipe page with working launch commands for serving V4 Flash across several hardware configurations, including the flag that turns on the DSpark speculative-decoding module and the FP8 KV-cache and expert-parallel settings DeepSeek recommends.
Desktop AI agent app
Berd →
Block's open-source Tauri desktop application for working with AI agents, wrapping the Goose backend over a WebSocket connection and adding projects, skills, extensions, automations, providers and session history in one window. Apache 2.0.
Desktop agent harness
Munder Difflin →
A local-first Electron app that runs a whole office of CLI coding agents on your own machine, wrapping 12 agent providers behind an on-disk message hive with per-agent inboxes and a single git committer to avoid lock collisions. Code and keys stay local by default.
Desktop assistant feature
ChatGPT Computer History →
An opt-in feature in the ChatGPT desktop app on macOS that turns activity across allowed apps and websites into a searchable timeline ChatGPT and Codex can reference, and can surface repeated workflows as suggested skills or automations. Off by default, no screenshots or audio, temporary event files deleted after 48 hours. OpenAI's own docs warn it increases prompt-injection risk.
Developer CLI
mcp-explorer →
Stateless command-line tool for probing any MCP server: list its tools, inspect a tool's input and output schemas, and call it with arguments. Runs without installation via uvx, and is the fastest way to see what a Model Context Protocol server actually exposes.
Developer tool
worktrunk v0.77.0 →
A command-line tool for managing Git worktrees, designed for running several AI coding agents in parallel on one repository. The 8 September release improves the wt list layout and makes JSON schema 2 the default.
Diffusion LLM API
Mercury 2 (Inception Labs) →
An API-only diffusion language model pitched on raw speed, claiming to out-pace open diffusion models on tokens-per-second for latency-sensitive generation.
Discovery benchmark environments
Apodex Discovery →
Executable environments built from real industry problems, with a rubric that scores an investigation's tools, repair, alternatives, coherence, evidence and scope independently of whether the final answer was right.
Distributed inference
Mesh LLM →
Runs models too big for one machine by splitting them across networked peers over serverless peer-to-peer transport; ~18 MB install, 40+ models up to 235B, OpenAI-compatible API on localhost.
Document OCR
Unlimited OCR →
Baidu's 3-billion-parameter document parser transcribes dozens of pages in a single pass without its memory footprint growing, because its decoder holds a constant-size cache instead of one that expands with every token. MIT licensed, with vLLM, ModelScope and ms-swift support already wired in, plus a hosted demo you can try in a browser.
Document reading (hosted)
Mistral OCR 4 →
A hosted document-reading model that converts scanned pages, PDFs, and complex layouts into clean structured text ready for a language model. Send a document, get back tidy text with the structure preserved.
Document workspace
Claude Docs →
Claude’s editable collaborative document surface with docx, PDF, Markdown and Google Docs export.
Document-to-text for AI
MinerU →
Open-source tool that converts complex PDFs and office files into clean markdown and structured data that AI models can read reliably. Run it yourself for free, with nothing leaving your machine.
Downloadable model weights
Qwen3.8-Flash-Next GGUF quants →
Unsloth's quantized builds of Qwen's newest architecture, in eleven sizes from roughly 72.5 GB at the smallest to about 354 GB at full precision. No official VRAM figure is published; community reports run a 4-bit build on a 16 GB card with around 100 GB of combined system memory.
Downloadable multimodal model
DeepSeek-V4-Flash-Vision-Exp →
An MIT-licensed 168 GB experimental multimodal V4-family checkpoint with a public model card and files.
E-ink app
riddle →
An open-source Rust app that turns a reMarkable Paper Pro into an interactive AI diary - handwrite a question and a vision LLM writes back in animated e-ink handwriting. Works with any OpenAI-compatible API. MIT-licensed.
Efficient open model
Nemotron-3-Puzzle-75B →
Nvidia's compressed 75B open model (from a 120B parent) with roughly double the serving throughput and 8x long-context concurrency on a single H100; weights on Hugging Face.
Embedded analytical database
DuckDB →
The in-process analytics database that a large share of data-science and AI-evaluation tooling runs on -- no server, just a library. Worth a mention today because AWS is acquiring DuckLabs while the project itself stays MIT-licensed under the nonprofit DuckDB Foundation, with more than a million downloads a day.
Embedded database
DoltLite →
A SQLite fork with Git-style version control over your tables -- branches, diffs, commits and merges on data. Reached beta on August 31, 2026; passes 100 percent of sqllogictest (5.8 million queries) and 99.46 percent of SQLite's acceptance tests.
Embedding model
NVIDIA Nemotron 3 Embed 8B →
8-billion-parameter retrieval encoder that turns queries and documents into normalized dense vectors for semantic search. NVIDIA claims state-of-the-art results on the multilingual RTEB leaderboard as of July 16; released under OpenMDW 1.1.
Encryption compiler
HEIR →
Google's compiler for fully homomorphic encryption: write a high-level program with annotations marking which values are secret, and it compiles down to backends including OpenFHE, Lattigo, tfhe-rs and Jaxite. Explicitly not an officially supported Google product.
Enterprise agent platform
Claude Tag (agent identity access model) →
Anthropic's product for putting Claude to work in shared team channels, now with an access model that gives each agent its own scoped accounts in the systems it touches -- GitHub, Slack, a data warehouse -- instead of borrowing an individual user's permissions, so every action is bounded and audited.
Evaluation harness
Claw-Eval →
Open benchmark for scoring multi-turn conversation quality in local models, separating answer quality from clarifying-question behavior. Used in TielCoder's published comparisons.
Experimental inference engine
WASTE →
A C inference engine that streams only the experts a mixture-of-experts model actually activates directly off NVMe, using spare RAM as an expert cache. Its author reports running the full 2.78-trillion-parameter Kimi K3 on a 64 GB laptop at about half a token per second. An existence proof, not a chat app.
Fast API language model
Mercury 2.5 →
Inception's diffusion language model, which refines a whole draft in parallel rather than writing left to right, reporting 1,107 tokens per second on standard NVIDIA GPUs with a 260K context window. Closed weights, available through Inception's API, Baseten and OpenRouter with 100 million free tokens.
Fast agentic LLM (API)
Gemini 3.6 Flash →
Google's newly GA fast model streams output nearly twice as fast as 3.5 Flash and costs less per task while holding the same intelligence-index score, tuned for high-volume agent loops that use fewer tokens and tool calls.
Fast speech synthesis
AuK-Flash →
The distilled sibling of AuK, generating in four steps instead of the full sampling schedule for markedly faster inference at the same 6.8 GB download size. Same MIT licence and the same instruction interface.
Find models & datasets
Hugging Face →
The main hub for finding, downloading, and trying open AI models and datasets — the field's town square.
Fine-tuning API
Tinker →
Thinking Machines Lab's hosted fine-tuning service, now serving Inkling alongside its other models. It is the managed path to customizing Inkling if you do not want to provision the GPUs yourself -- with the caveat that the API caps context at 256K tokens, versus 1M for the open weights you run yourself.
Fine-tuning toolkit
Unsloth (AMD support) →
The fine-tuning and RL toolkit now documents AMD support across training, RL, chat, and deployment on Windows, WSL, and Linux, plus a cross-platform Studio beta.
Firmware security platform
Exein Analyzer →
A production product for firmware and SBOM analysis, reachability-based vulnerability prioritization and AI-assisted remediation workflows.
Forecasting model
TimesFM →
Google's pre-trained foundation model for time-series forecasting — predicting things that change over time, like demand, traffic, or sensor readings — usable out of the box without training your own model.
TimesFM 3 →
Google's 330M-parameter multivariate zero-shot forecasting model, downloadable for non-commercial, non-production use.
Form-filling model
CUA-S1-FORMS →
An MIT-licensed specialist that selects candidate document values and fixed actions for structured form interactions.
Formal mathematics library
Formal Conjectures →
Google DeepMind's open Lean library of formally stated open mathematical conjectures, now the venue where the claimed Jacobian conjecture counterexample is being reviewed in public. A usable resource if you want machine-checkable statements of open problems rather than prose.
Formal methods
Anthropic Fermat's Last Theorem Lean repository →
A public Lean codebase for Anthropic's formalization of a classical proof route for Fermat's Last Theorem.
Formal verification for neural networks
TorchLean →
A Lean 4 framework for formalizing, executing and verifying neural networks, with typed tensors, exact and finite-precision semantics, verified reverse-mode differentiation, and CROWN/LiRPA-style bound checking. Early and CPU-bound by its authors' own account, but it is the most concrete attempt yet at machine-checked robustness guarantees.
Free hosted model
Ox Alpha on OpenRouter →
A free anonymous stealth model with a 1,048,576-token context window, up to 131,072 output tokens, and text, image, and video input. Genuinely usable and genuinely free right now, with a caveat worth reading first: OpenRouter states it is not the developer or provider, its model page says prompts and completions are retained by the anonymous provider, and other documentation for the same model claims zero retention.
Frontier chat & coding model
Grok 4.5 →
SpaceXAI's new 1.5-trillion-parameter model, available in Grok Build, Cursor, and the API at $2 per million input / $6 per million output tokens, with a full public release on July 9.
Frontier model
Claude Fable 5 (redeployed) →
Anthropic's top-tier model, back online after a brief export-control suspension, now shipping with a hardened cybersecurity classifier that reroutes flagged requests to Opus 4.8 and a wider default safety margin.
Frontier model (API)
Claude Fable 5.1 →
Anthropic's new generally available model for long-running agentic coding and knowledge work, live on the Claude API as claude-fable-5-1 and on AWS, Google Cloud and Azure. Cache reads dropped 75 percent to $0.25 per million tokens; base rates unchanged at $10 in and $50 out. Now permitted to find software vulnerabilities in source code.
Frontier model API
GPT-5.6 (Sol / Terra / Luna) →
OpenAI's newest model family, tuned for cheap, fast, reliable agentic work, with programmatic tool calling, a multi-agent beta, persisted reasoning, and a high-reliability 'pro' mode.
GPT-6 Astra API →
OpenAI's agentic flagship, aimed at computer use, browsing, coding and long multi-step workflows. Five reasoning effort levels from low to max, with no off switch. $10 per million input tokens and $50 per million output, cached input at $1 -- cache discipline is the difference between an affordable agent loop and an unaffordable one.
GPU kernel SDK
cuTile Rust →
Experimental Rust tile-kernel toolkit that JIT-compiles through CUDA Tile IR.
cuda-oxide →
Early-alpha Rust frontend for explicit CUDA SIMT kernel authoring.
GPU kernel verification
FpSan (Floating-Point Sanitizer) →
Open-source correctness checker for Triton GPU kernels, and the tool OpenAI says it used to validate the production kernels GPT-5.6 Sol rewrote. It compares symbolic computation under its own payload algebra rather than simulating IEEE floating point, so results should be compared only against other FpSan runs. Useful for anyone writing or generating custom kernels who needs to catch numerical breakage before it reaches production.
GUI-agent benchmark
Pointer Bench →
A 1,500-task benchmark for GUI grounding across spreadsheets, text documents and professional applications, built by Warmwind because general agent scores do not transfer to office software. Public leaderboard, open dataset on Hugging Face, and code on GitHub -- useful if you are evaluating whether a screen-driving agent can actually hit the right cell.
Genome language model
Evo →
Arc Institute's family of genome language models, released openly with code and checkpoints. Used by Arc and Stanford to generate complete synthetic bacteriophage genomes that were then built and tested in the lab against non-pathogenic bacterial hosts.
Genome model
Evo 2 →
Arc Institute's open genome language model for DNA, used to design bacteriophage genomes that were synthesized and shown to work in living bacteria; weights and code are public.
Genomics database
AlphaGenome Atlas →
Free browsable database of predicted molecular effects for all 9 billion possible single-letter changes to human DNA, with a single impact score per variant. Free for academic and non-commercial use via web portal and API; explicitly not validated for clinical use.
Give AI agents code memory
codebase-memory-mcp →
Indexes an entire codebase into a persistent, queryable knowledge graph so AI agents can understand large projects fast. Supports a huge range of programming languages, answers queries near-instantly, and ships as a single dependency-free binary.
Hardware peripheral
Codex Micro →
A $230 mechanical control deck for driving OpenAI's Codex agents, built with keyboard maker Work Louder. 13 switches, a joystick, a touch sensor, RGB keys showing live agent status, and a rotary dial that adjusts reasoning effort -- turning an API parameter into a physical knob. Nothing it does is impossible with keyboard shortcuts; the pitch is ambient awareness when supervising several agents at once.
Hosted LLM API
DeepSeek V4 Pro (API) →
A strong open-weight reasoning and coding model now offered through DeepSeek's own API at a permanently cut, low per-token price, undercutting frontier closed models for high-volume work.
Hosted demo
AuK Demo Space →
A browser demo of AuK on Hugging Face Spaces for judging the output quality directly without downloading 6.8 GB of weights first, which for a generative audio model is the only assessment that counts.
Galahad verified-reuse testbench →
Public testbench for the frozen-12B verified procedure cache, where a solved and independently verified problem family is answered on later instances at zero generation tokens, bit-exact. Worth poking at to understand what the claim does and does not cover -- the engine source, configuration and raw artifacts are withheld, so this demo plus the bench repo is the only inspectable surface.
Hosted inference API
Cerebras Inference (Qwen 3.8 27B) →
Serves the open Qwen 3.8 27B at roughly 1,500 output tokens per second, with a free tier at 64k context and paid at 128k. Automatic prompt caching cuts time-to-first-token. Read the rate limits first -- the free tier's 90,000 tokens per minute lands almost exactly at the model's own output rate.
Hosted model
GLM-5.3 on OpenRouter →
Z.ai's GLM-5.3 with a 1 million token context window and always-on reasoning, billed at $1.40 per million input tokens and $4.40 per million output, with cheaper cache reads. Tuned for long-horizon software engineering and vulnerability discovery.
Hosted model API
DeepSeek V4 Flash 0731 (API) →
The updated V4 Flash checkpoint now serves behind the existing deepseek-v4-flash identifier, with a 1-million-token context, 384K maximum output, tool calls, and an OpenAI-, Anthropic- and Responses-API-compatible interface. Fresh input runs $0.14 per million tokens, output $0.28, and cached input $0.0028 - a fiftyfold discount on repeated prefixes.
Gemini 3.6 Flash and 3.5 Flash-Lite →
Google's economy-tier models went generally available on July 21, with 3.6 Flash keeping a million-token context and 64,000-token output while dropping its output price roughly a sixth versus 3.5 Flash and using about 17 percent fewer output tokens per task. Note the migration-breaking changes: some sampling parameters are deprecated and prefilled model turns are no longer supported.
Gemini 3.7 Flash →
Google's cheap workhorse tier, now aimed squarely at coding and agents, with a 1,048,576-token input window and 65,536-token output. Introductory pricing of $0.75 per million input tokens and $3.75 output runs through December 31, 2026, after which the rate doubles.
Grok 4.6 →
xAI's new frontier model, tuned for long-running agents and available day one in Cursor, Grok Build, and the xAI API. Two dollars per million input tokens and six per million output, with a faster variant at double the price.
Qwen3.8-Max →
Alibaba's new flagship multimodal model, live today as a paid API at $2 per million input tokens and $6 per million output tokens, with a one-million-token context, function calling, structured output, and prompt caching that drops repeated input to $0.25 per million. Weights are promised but not published.
Seed2.0 (ByteDance Seed) →
ByteDance's Seed2.0 family (Pro, Lite, Mini) of closed, API-hosted models aimed at long-tail knowledge and complex instruction-following, accessed through ByteDance's Volcano Engine (Ark) platform. Not open weights despite the academic-style model card.
Hosted open-model API
GLM-5.2 on Baseten →
The top trending open-weight model served as a fast hosted endpoint, reported at 280+ tokens/sec on Blackwell-class hardware -- an open model you can call like a closed one.
Hosted video generator
Hailuo AI Video →
MiniMax's hosted front end for H3, for trying the model in a browser before downloading tens of gigabytes of weights. Supports text-to-video, image-to-video, first-and-last-frame and reference-to-video workflows.
Image API
GPT-Image-2.5 Flare and Sunburst →
Two new OpenAI API image models: Flare for high-volume generation where speed matters, Sunburst for detailed creative work needing extra precision at the cost of longer generation times.
Image generation
ChatGPT Images 2.5 (with Sketch) →
OpenAI's updated image model with up to 50% lower generation latency, better preservation of subjects from reference photos, and Sketch, a tool for drawing directly in ChatGPT as a reference. Available to all ChatGPT, ChatGPT Work and Codex users.
Mage-Flow →
A Microsoft demo space for image generation and editing that works at native resolution rather than upscaling from a fixed square, running free on Hugging Face's shared GPU tier.
Muse Image →
Meta's agentic image model, free for everyday creation inside Meta AI, Instagram Stories (US), and WhatsApp; it can search, write code, and self-refine rather than mapping a prompt straight to pixels, and stamps outputs with an invisible Content Seal watermark.
Image generation API
Nano Banana 2 Lite →
Google's fastest, cheapest Gemini image model - a text-to-image picture in about four seconds for roughly three cents per thousand images, built for high-volume use.
Image generation and editing
Boogu-Image 0.1 →
An open-source unified image understanding and generation model family (Base, Turbo, Edit, Edit-Turbo) with instruction-based editing and bilingual Chinese-English text rendering, trained for roughly $400K. Apache 2.0.
SenseNova-U1.5-8B-MoT →
Apache 2.0 model that generates and edits images without a vision encoder or latent autoencoder, working on pixels directly. Handles natural-language edits, multi-image references, insertion and replacement, and region control via bounding boxes; ships quantized and offload paths for 24 GB-class GPUs.
Image, video and audio generation
FLUX 3 (early access) →
Black Forest Labs' unified generation model, producing video up to 20 seconds with native synchronized audio from text, image, video or keyframe inputs. Video is behind an early-access request today; image access is promised in the following weeks and open weights are deferred.
Independent benchmarking
Artificial Analysis model pages →
Third-party cost and capability measurements for frontier models, including the cost-per-completed-task figures that contradicted Anthropic's own pricing framing for Fable 5.1 on launch day. The most useful free counterweight to vendor benchmark tables.
Inference acceleration
DFlash 2 (Qwen3.8-27B drafter) →
A drop-in block-diffusion drafter for speculative decoding on Qwen3.8-27B, with documented launch commands for SGLang and vLLM. Output is provably identical to the target model; throughput gains reach 3.4x on single requests and shrink under heavy concurrency.
Sol-Attn (Sol-Engine) →
NVIDIA's drop-in sparse attention kernel for long-video diffusion transformers, released July 28 for HunyuanVideo-13B and Wan2.1-T2V-14B. Screens compressed key/value blocks inside a single online-softmax pass, so exact attention goes where it matters and skipped blocks get an approximate correction. Training-free, no weight changes, reported up to 2.1x for generation and 2.3x for editing. The repo marks end-to-end re-benchmarks for the two integrated pipelines as pending.
Inference engine
SGLang (Kimi K3 cookbook) →
Alternative open-source serving engine with day-zero K3 support and a step-by-step deployment cookbook. Its writeup documents how prefix caching, paging and prefill/decode disaggregation were rebuilt to handle K3's mix of recurrent and key-value state.
Inference optimization
Gambit →
An inference algorithm that prunes unpromising reasoning trajectories and immediately branches new ones from strong prefixes, keeping the hardware busy. Its authors report up to 68.5 percent fewer total tokens than standard parallel sampling with higher accuracy; code is public.
Inference platform
Doubleword (async + batch inference) →
Run the same models you already use, but on async and batch tiers that trade latency for a large cost cut on workloads that don't need an instant reply: long-running agents, evaluations, and bulk jobs.
Inference runtime
LvLLM →
A community inference runtime specialised in hybrid CPU and GPU execution of mixture-of-experts models, with NUMA-aware scheduling, expert weight management, and MXFP4 quantization kernels. Its DeepSeek V4 build publishes a working dual-RTX-3090 launch configuration at 22K context.
Inference server
vLLM (Kimi K3 support) →
The widely used open-source serving engine landed day-zero Kimi K3 support with a documented recipe, an FAQ on minimum hardware, and a K3-specific DSpark draft model for speculative decoding that roughly triples single-user throughput.
vLLM DeepSeek-V4 support →
vLLM shipped serving support for DeepSeek-V4's compressed long-context attention, including hybrid KV-cache management, multiple cache page sizes, kernel fusion and multi-stream partitioning. The engineering post documents the recipe and the hardware it assumes.
Interactive AI broadcast platform
fal.live →
A public platform for continuous AI-generated broadcasts where viewers submit and vote on what happens in the next scene.
Interactive demo
Neuronpedia J-lens demo →
A live, no-install web demo of the Jacobian lens that lets you watch the 'contents of the workspace' light up inside open models (Qwen 3.6 27B and Gemma 3 12B) as they process text.
Interactive video service
Vidu Stream →
A public Vidu service surface for real-time generative video experiences, including the vendor’s S2-era interactive workflows.
Interactive world model
AlayaWorld →
Inference code and pretrained weights for an autoregressive world model with real-time camera control, prompt switching and long-horizon memory consistency, using an explicit 3D cache for spatial recall plus a compressed frame-history embedding. Training code is not included and the weights ship under a community license.
Evoke →
Open-weights 14B world model that generates a navigable video world you can steer with a camera and text mid-session, keeping scene geometry in an external memory bank so places stay consistent when you look back. Apache 2.0, with every training-stage checkpoint published, not just the final one.
Interpretability tool
Jacobian Lens (J-lens) →
Anthropic's open-source tool that reads a model's silent 'working memory' - for any word, it finds the internal pattern that makes the model more likely to say it later. Apache-2.0, with a live interactive demo on open models.
jlens-gguf →
A GGUF-native implementation of Anthropic's Jacobian Lens for local models, with a browser UI to visualize, swap, and ablate a model's internal concepts live as it generates through llama.cpp.
Interpretability visualizer
chessformer_lens →
A pip-installable mechanistic interpretability lens for square-token chess transformers. It visualizes the move policy live and lets you ablate any attention head with a click, on top of the Maia-3 model family.
Knowledge-work connector
Notion hosted MCP server →
Notion’s OAuth-backed hosted Model Context Protocol server for searching, reading and updating authorised workspace content from compatible AI clients.
LLM API and open weights
DeepSeek V4 →
DeepSeek's latest model family (a 1.6T-parameter Pro and a 284B Flash, both with a 1-million-token context by default), available as an API and as open weights on Hugging Face.
LLM API router
Dahl Inference →
Third-party inference router reselling top open-weight models (Kimi K2.6, MiniMax M2.7, GLM 5.2) at low per-token prices, currently running a 100M-free-token promotion.
LLM chat and API
Kimi K3 →
Moonshot AI's 2.8-trillion-parameter flagship with a 1M-token context window, tuned for agentic coding and knowledge work; it topped a frontend-coding leaderboard. Usable now via kimi.com chat and an OpenAI-compatible API, with open weights due July 27.
LLM cost routing
World Model Optimizer →
A pip-installable CLI that turns the OpenTelemetry traces your agents already emit into a routing policy: it scores every model you have registered against held-out tasks from your own traffic, then serves an endpoint that sends easy requests to cheap models. Treat the routing as the product; the distillation half has no released checkpoint yet.
LLM inference acceleration
DeepSeek DSpark →
Open-source speculative-decoding implementation using parallel tree drafting to speed up text generation with no change to the model's output - the project that topped Hacker News this week. Drop-in inference speedups for self-hosted models.
JetSpec →
Parallel tree-drafting speculative decoding aiming for large, lossless inference speedups; project page and writeup with code, reporting up to several-times faster generation depending on the model and workload.
LLM security scanner
garak v0.17.0 →
NVIDIA's open-source vulnerability scanner for language models, released 9 September. This version maps probe results to EU AI Act risk categories and adds authentication options for Ollama. Apache 2.0.
LLM-text detection for crowd data
GPTurk →
EPFL's released code for detecting language-model-assisted submissions in crowd work, combining keystroke logging with a synthetic-versus-real text classifier. Practical for anyone buying human-labeled data who needs to check whether the labels were actually written by people rather than pasted from a chatbot.
Lab automation
Opentrons Python API →
The mature, vendor-supported way to script a liquid-handling robot today: a documented Python and HTTP interface for the Flex and OT-2 platforms, pipettes and modules. Worth knowing as the existing baseline that Anthropic's new hardware standard is being measured against.
Lab instrument standard
SiLA 2 →
A free and open standard for laboratory instrument interoperability built on HTTP/2 and Protocol Buffers, with a multi-part specification and a public repository. The incumbent open standard in the space AI-native hardware interfaces are now entering.
Language model API
Ling-3.0-flash (free API) →
Ant's 124B-parameter mixture-of-experts model that activates only 5.1B parameters per token, with a 256K context and OpenAI- and Anthropic-compatible endpoints. Currently free on OpenRouter as inclusionai/ling-3.0-flash:free; aimed at long-horizon agent workflows and tool calling.
Lean proof repository
OpenAI NavierStokesAndEuler →
The Apache-2.0 Lean 4 formalization accompanying OpenAI's Navier-Stokes and Euler blowup claims, building against Mathlib and including a directory set up for independent proof-checking.
Live video generation API
fal H3 Max Director →
fal's API for continuous AI-generated video streams with live prompts and chunked playback controls, built around MiniMax H3 Max Director.
Local AI runtime
Ollama 0.31 →
Run open models on your own computer; the new version nearly doubles Gemma's speed on Apple Silicon using multi-token prediction, on by default.
Local CPU model
SHADOW-50M →
A proof-of-concept 44M-parameter CPU model with exact arithmetic circuits, disk-backed retrieval and WebAssembly support.
Local coding model
Gemma-4 12B Coder (GGUF) →
A fine-tuned, locally-runnable version of Google's Gemma-4 model specialized for programming tasks, packaged in a format that runs efficiently on everyday consumer hardware.
Ornith-1.0-9B GGUF →
The quantized build of the smallest member of the MIT-licensed Ornith-1.0 family, an open agentic coding line post-trained on top of Gemma 4 and Qwen 3.5. The Q4_K_M file is 5.63 gigabytes, which puts it within reach of a single consumer GPU.
TielCoder 35B-A3B GGUF →
4-bit dynamic re-quantization of Ornith-1.5-35B-A3B for llama.cpp. The benchmarked 22.4 GB tier fits a 24 GB card and fixed 12 of 25 live software issues in the maintainer's tests. Includes a vision projector for reading screenshots and stack traces.
Local coding-agent model
Poolside Laguna S 2.1 →
A public-weight, 118B-total mixture-of-experts coding model with only ~8B active parameters that runs locally on a single 128GB machine via a 75GB Q4 GGUF, built for long-horizon agentic software work under the permissive OpenMDW-1.1 license.
Local decision model
Von 1.0 →
An Apache-2.0 395M ModernBERT-based model and runtime for local structured decisions and HTTP serving.
Local fine-tuning desktop app
Unsloth Studio →
A no-code desktop and web interface for training and running language models on your own machine, with a one-line installer and desktop shortcuts. Training is NVIDIA-GPU centric; inference and export work across Mac, Windows and Linux.
Local image studio
Qwen-Image-2.1 →
Downloadable image-generation and editing pipeline with reference images, masks and native transparent-layer support; the core files are about 33.1 GB and commercial use requires a separate licence.
Local inference
TurboFieldfare →
A Swift and Metal runtime that runs Gemma 4's 26B model on an 8GB MacBook Air by keeping a 1.35GB core resident and streaming the rest of the experts off the SSD. Ships as a Mac app, a CLI and an OpenAI-compatible local server.
Local inference build
llama.cpp-gfx906 →
A llama.cpp fork with hand-written kernels for AMD's GFX906 architecture, making used Instinct MI50, MI60, and Radeon VII cards usable for local inference. Ships custom flash-attention, RoPE, and matrix-multiply paths plus overclocking and power-scaling scripts.
Local inference engine
Deltafin →
A Rust engine that runs the full uncompressed 2.8-trillion-parameter Kimi K3 on a single MacBook Pro by streaming expert weights from external SSDs, sustaining about one token per second with nothing quantized or pruned. A demonstration of the ceiling on local inference rather than a daily driver.
bitnet.cpp →
Microsoft's official inference framework for 1.58-bit ternary language models, built on llama.cpp with optimized CPU and GPU kernels for running very heavily compressed models on ordinary hardware.
llama.cpp b10228 →
The release that adds DeepSeek V4 Flash's embedded DSpark speculative-decoding head, plus a converter that can split the draft tensors into a separate GGUF. Gains are workload-dependent: roughly 2x decode on large multi-GPU setups, and a measured slowdown on a 24 GB card with CPU offload.
llama.cpp v0.1.0 →
The engine behind most local AI setups published its first semantic-looking version tag on August 17, 2026, pinned to commit 7c35571. Useful mainly to packagers and anyone who needs a version string a dependency resolver understands; the project still makes no API-stability promise.
Local inference runtime
CachyLLama →
MIT-licensed llama.cpp fork that saves conversation and system-prompt caches to SSD and restores them after a restart, so local agents stop reprocessing the same prompt prefix every turn. Its own benchmark reports long repeated agent prefixes going from minutes cold to about a second warm.
FastFlowLM →
An NPU-first, GPU-free inference runtime built exclusively for AMD Ryzen AI (XDNA) NPUs, targeting long-context local LLMs at low power on laptop-class hardware; the team just joined AMD, with open install guides for Ubuntu, Arch, and more.
llama.cpp b10217 →
The 1 August build adds support for DeepSeek V4 Flash emitting tool calls inside its reasoning block, which is what was silently killing local agent runs against the new model. If you are running DS4 locally with tools, this is the build you need.
Local inference server
llama.cpp (MCP tool hosting) →
The most widely used local LLM server now launches and manages local Model Context Protocol tool processes itself, discovers their tools and exposes them through its chat API - turning a plain inference server into an agent host. Off by default; needs a tool-capable chat template.
Local model
Bonsai 2 →
Prism's downloadable ternary 27B model and runtime for local text, tool-calling and optional vision workflows.
Local model package
Maple-Preview (2-bit MLX build) →
DeepGrove's 20B mixture-of-experts model with about 1B active parameters per token, packaged for Apple Silicon at roughly 5.3GB. The build uses affine two-bit group quantisation with four-bit embeddings and output head, and its loader packs ternary values into two-bit codes. Note that the native BF16 repository is about 40.4GB, and DeepGrove publishes no ternary training recipe or independent evaluation.
Local model runner
slotstream →
A single Swift binary that runs the 104 GB Qwen3.8-Flash-Next mixture-of-experts model on Apple Silicon Macs with far less memory, by streaming expert weights off the SSD. Speaks the Ollama and OpenAI chat APIs, so existing tools work unchanged. About 12 tokens per second on a 48 GB Mac; needs roughly 110 GB of free disk.
Local model runner / fine-tuning
Unsloth →
Toolkit and documentation for running and fine-tuning large open models faster and on smaller hardware, including aggressive dynamic quantization recipes that shrink models like GLM 5.2 by 80-plus percent while keeping most of their accuracy. The practical on-ramp to running near-frontier models privately.
Local model runtime
SwiftLM →
An MLX-based runtime for Apple silicon whose --stream-experts mode reads mixture-of-experts weights straight off an NVMe SSD, letting a machine run models several times larger than its RAM.
Local model server
KoboldCpp v1.118 →
Single-binary local model server that shipped its own fix for multi-turn DeepSeek V4 Flash prompt-processing problems on the same day as the llama.cpp fix. Useful if you want a working DS4 setup without building anything.
Local model weights
DeepSeek-V4-Flash-0731 GGUF (Unsloth) →
Community quantizations of the new MIT-licensed DeepSeek weights in GGUF form, running from roughly 83GB at aggressive low precision to about 162GB at 8-bit. Usable on high-memory workstations and multi-GPU rigs, not on a laptop.
Qwen3.8-27B GGUF builds →
Ready-to-run compressed builds of Alibaba's 27-billion-parameter multimodal Qwen3.8, covering the full ladder from eight-bit down to one-bit. Community testing points to the six-bit build, around 22 gigabytes, as the conservative floor for serious agentic coding, with three-bit still usable and one-bit rebuilds degrading sharply because the calibration file has no data for the model's multi-token-prediction head.
Unsloth Qwen3.8-27B GGUF →
Quantized builds of Alibaba's newest 27B open-weight model, published within minutes of the release, in a range of sizes that fit on a single consumer graphics card.
Local video generation front end
h3-studio →
Local front end for MiniMax H3 built on ComfyUI 0.30.0 or newer, with a VRAM meter, idle GPU release and a service setup so the video model hands the card back to other workloads. Its notes document H3 needing roughly 15.5 GB in the NVFP4 build, which is the practical ceiling on a 16 GB card.
Local voice AI toolkit
Voicebox →
A local-first voice stack bundling voice cloning, TTS, Whisper transcription and dictation, a refinement model, a REST API, and a built-in MCP server so an agent can speak, transcribe, and manage voice profiles without cloud calls.
Low-bit language model
Bonsai 27B (GGUF) →
PrismML's roughly 27.8-billion-parameter Qwen-derived model trained with 1-bit binary or 1.58-bit ternary weights end to end, which the company says fits in about 4 GB and runs on phone-class hardware. Performance figures are vendor-reported and not independently replicated.
MCP app framework
Skybridge →
A framework for building MCP-native apps -- interactive tools an AI assistant can open and use directly, pitched as 'MCP apps are the new website.'
MCP server/client toolkit
FastMCP →
A Python toolkit that turns ordinary functions into Model Context Protocol tools, resources, and prompts with generated schemas, validation, and docs, and a client that handles transport negotiation, auth, and protocol lifecycle.
Managed model API
Qwen3.8-Max-0902 →
Alibaba Cloud's 2.4T-parameter MoE flagship with native vision, long-horizon-task support, and a one-million-token context window.
Market data dashboard
Ramp AI Index →
A free public dashboard tracking AI vendor adoption across US businesses, derived from corporate card and invoice payments covering more than 100 billion dollars in annual spend across over 50,000 companies. Useful as an adoption-behaviour signal, with the important caveat, stated in Ramp's own methodology, that it measures spend rather than vendor revenue.
Media provenance tooling
remove-ai-watermarks →
A Python library and CLI that detects and removes visible marks, invisible watermarks such as SynthID, and provenance metadata including C2PA, EXIF, IPTC and XMP from images and video. Useful for defenders auditing how durable their own provenance actually is.
Medical imaging research
RADAR →
A public research release for scoring broad abdominal-CT findings from anatomy-aware vision-language representations; not a clinical deployment product.
Minimal agent harness
minion →
Harrison Kinsley's deliberately lightweight coding harness, used as the control in his local benchmarks. Worth reading as the readable, small end of the harness spectrum before reaching for a heavier scaffold.
Mobile assistant
Siri AI in iOS 27 →
Apple’s shipping iOS 27 assistant offers conversational requests, personal-context search, app actions, web answers and visual-intelligence features, subject to Apple Intelligence device and regional availability.
Model API
Claude Opus 5.5 →
Anthropic's latest high-capability Claude tier for complex coding and agentic tasks, with configurable effort for cost-quality trade-offs.
CompactifAI Quasar 438B →
A served 438B compressed coding-model endpoint with chat completions, tool calling, and structured output; public downloadable weights were not verified.
GPT-6 Luna →
OpenAI's inexpensive high-volume model for focused workflows, with text/image input, Responses API support and a 1.05M-token context window.
GPT-6 Sol →
OpenAI's lower-cost model for difficult coding and agentic work, with text/image input, Responses API support and a 1.05M-token context window.
Grok 4.7 API →
xAI's 500K-context reasoning model for coding, agentic tasks, and knowledge work.
Meta Model API (Muse Spark 1.1) →
Meta's first paid, hosted model API, built around the Muse Spark 1.1 multimodal reasoning model -- a million-token context window with active context compaction, zero-shot tool and MCP support, and an OpenAI-compatible interface so existing code drops in with little more than an endpoint change.
Model benchmark index
Vals AI →
Independent evaluator that scores frontier and open-weight models on professional workloads - finance, tax, legal, medical, public benefits - alongside coding benchmarks, with per-model cost figures. A useful counterweight to vendor-published charts.
Model code and prompting guides
MiniMax H3 GitHub →
Official repository for running H3 locally, including inference code and MiniMax's own prompt-writing skills for getting usable results out of the multimodal context format.
Model comparison
Artificial Analysis Intelligence Index →
The independent benchmark and pricing dashboard the field now reaches for when a lab claims a lead -- it is the source of Inkling's debut score of 41. Useful beyond the headline ranking because it also tracks output tokens per task, latency and cost, which is how you find out that a cheaper-looking model is actually more expensive per finished job.
Model comparison and benchmarking
Artificial Analysis →
A free public dashboard that independently benchmarks and compares AI models on a combined intelligence index alongside price and speed; the source of this week's finding that GLM-5.2 leads the open-weight class.
Model control heads
Multi-Head Latent Control →
Freezes a model and attaches two small heads that read its hidden states to decide whether to answer, use a tool, ask for information, abstain, or escalate to a stronger model. Open-sourced with matching small checkpoints; needs white-box access.
Model endpoint
TypeSafe Jev on Cloudflare →
Cloudflare documents a live typesafe/jev third-party model surface for bounded choice, score, and Boolean decisions.
Model fingerprinting toolkit
ox-alpha identification harness →
A working, dependency-light harness for identifying an anonymous model endpoint: an interactive multi-turn CLI plus a parallel probe runner that logs every raw request and response. Its tokenizer-differential technique - comparing reported prompt-token counts for the same string across models - identifies a model family without needing the model to cooperate, and it runs against any OpenAI-compatible API.
Model gateway
OpenRouter →
A production gateway to hundreds of models behind one API, with public rankings built from real usage and the ability to sort by price, throughput, latency and popularity.
OpenRouter programming collection →
OpenRouter's curated collection of models for coding and agentic work, including the stealth/ox-alpha listing. Useful for A/B testing several coding models behind one API without separate accounts, and for checking a model's advertised context and modality contract before you build against it.
Vercel AI Gateway (Ling-3.0-flash) →
Vercel added Ling-3.0-flash to its AI Gateway with bring-your-own-key support and failover routing, free through August 3. Useful if you want the model behind a single gateway alongside other providers rather than wiring a second API.
Model integrity auditing
abliterlitics →
An evaluation harness for checking whether an edited or guardrail-stripped model is actually intact: it diffs every tensor against the base model, measures behavioural drift on harmless prompts, runs a multi-domain capability suite, and scores harmful-completion rates separately. A tensor diff from this would have caught this week's broken Gemma 4 release in seconds.
Model marketplace
OpenRouter discounted models →
A live collection of models currently carrying provider discounts on OpenRouter. GPT-5.6 Sol from the OpenAI provider is listed at roughly half OpenAI's own promotional rate, against $5 and $30 for the same model via Azure.
Model optimization toolkit
Voodoo Dynamic Quant →
MIT-licensed tooling that learns per-tensor mixed-precision allocations and exports llama.cpp-compatible GGUF quantizations.
Model orchestration API
Sakana AI Fugu Max and Fugu Ultra v2 →
Orchestration models that route each task across a pool of other models, available from 11 September through an OpenAI-compatible API. Fugu Max is priced at $2 per million input tokens and $6 per million output tokens; existing Fugu users upgrade with a one-line parameter change.
Sakana Fugu Max →
Launched 11 September: a model trained to split each request across a pool of other models and combine the answers, sold as one OpenAI-compatible API at $2 per million input tokens and $6 per million output tokens, with a 1 million token context. The pool is not published beyond NVIDIA's Nemotron family, all quality claims are Sakana's own, and it is not yet available in the EU or EEA. Listed on OpenRouter and Vercel AI Gateway.
Sakana Fugu Ultra v2 →
The higher-priced sibling of Fugu Max, also released 11 September, at $5 per million input tokens and $30 per million output tokens (rising above 272,000 tokens of context). Sakana says Claude Fable 5, Fable 5.1 and GPT-6 Astra are not in its model pool. Available through OpenRouter as sakana/fugu-ultra-v2.
Model playground and API
Gemini 3.8 Flash in Google AI Studio →
Google's newest Flash-tier model, aimed at long-horizon coding and agent work, with a one-million-token context window and 64,000-token output. Free to try in AI Studio; API pricing is $0.75 per million input tokens and $3.75 output through the end of 2026. It deliberately spends more tokens on hard tasks, so budget by cost per finished job rather than per token.
Model preservation
Pirate Face →
Public model metadata and torrent index that can use Hugging Face web seeds and peer seeding to preserve exact model files, subject to licence, safety and availability caveats.
Model proxy
Free Claude Code →
MIT-licensed local proxy that lets Claude Code, Codex, OpenCode and other coding agents run against roughly 50 different providers, preserving Anthropic's wire protocol so the client never notices. Routes each internal model tier to a different upstream.
Model quantization
Unsloth Kimi-K3-GGUF →
Converted local-inference builds of Moonshot's Kimi K3: a 1.51 TB four-bit UD-Q4_K_XL file, a 1.56 TB eight-bit build, and BF16/F16/F32 multimodal projector files that preserve an image-input path. Datacenter-scale hardware still required.
Voodoo Quant →
A per-tensor sensitivity-aware quantization method that spends more bits on important tensors, claiming large divergence reductions over standard llama.cpp and Unsloth quants, especially at 1-bit and 2-bit; GGUF files run in unmodified llama.cpp.
Model routing
OpenRouter Auto router →
Single endpoint that picks a model per request using the past seven days of aggregate platform spend on similar tasks, with a cost_tier parameter to set how much you want to spend and account-level guardrails respected.
Model routing infrastructure
LLMRouter →
A unified framework for building, evaluating and deploying model routers, with a quickstart, single and batch routing calls, and a benchmark that dispatches queries across eighteen candidate models with cost tracking.
Model routing library
NeMo Switchyard →
NVIDIA's library for routing each task in a multi-model system to the model best suited to it, so a frontier model handles planning while a cheaper one handles execution. Shipped alongside Nemotron 3.5 Lightning as the connective tissue for mixed-model agent stacks.
Model routing quality control
OpenRouter Auto Exacto →
OpenRouter's provider-routing system that repeatedly evaluates provider telemetry and benchmark behavior, then deranks statistical outliers.
Model serving engine
SGLang →
The other serving stack DeepSeek's official 0731 model card documents as supporting DSpark directly, alongside the recommended FP8 key-value cache and FP4 indexer cache configuration for V4 Flash.
Model weights
Inkling-Small GGUF →
Quantized builds of Thinking Machines' newly released 276B/12B multimodal open-weight model, packaged for llama.cpp, LM Studio and Ollama so you do not have to download the 532GB original.
Model-orchestration API
Sakana Fugu →
A single OpenAI-compatible endpoint that dynamically routes each request across several frontier models, so you call one API and get a coordinated multi-model answer.
Monocular depth estimation
Depth-Anything-3 →
ByteDance's depth estimation model and code, used as a geometry backbone by other systems including AlayaWorld. Weights are published on Hugging Face and the repository is the standard integration path for recovering per-pixel depth from ordinary images and video.
Multi-agent framework
LatentMAS →
A training-free framework for multi-agent collaboration that passes last-layer hidden states and cached internal state between agents instead of text messages, reporting 70.8 to 83.7 percent fewer output tokens and roughly four times faster end-to-end inference. Already has an extension ecosystem including science, retrieval and hybrid variants.
agency-agents →
An open-source library of 150-plus specialized AI agent personas across 13-plus professional divisions, built to run multi-agent workflows natively in Claude Code with conversion scripts for other agentic coding tools.
Multi-agent research environment
Station →
An open-source open-world environment where AI agents from different model families pursue a shared research goal with no coordinator, choosing directions and writing into a shared literature. Suited to tasks that are scorable and finish in about two hours. Needs model-provider API keys and the OpenAI Codex CLI.
Multi-agent research framework
Virtual Lab →
The open-source multi-agent research framework behind the Nature nanobody paper, where an LLM principal investigator coordinates specialist agents over tools like ESM, AlphaFold-Multimer and Rosetta. Runnable on your own project with your own agent roster.
Multi-agent research system
SearchOS →
Open-source (MIT) multi-agent web-research framework that treats search like an operating system: progress lives in an explicit evidence graph, coverage map, frontier task queue, and failure memory instead of chat history, with a pipeline-parallel scheduler. Ships a CLI/TUI, web frontend, installer, and replayable sessions.
Multimodal API
Qwen3.8-Omni-Flash →
Alibaba's hosted API for long-context text, image, audio and video analysis with function calling and web search.
Multimodal API model
DeepSeek V4 Flash Vision (experimental) →
An experimental multimodal version of DeepSeek's cheapest model, live on the DeepSeek API as deepseek-v4-flash-vision-exp. It takes images inline with text via base64, external URL, or the Files API, budgets each image to at most 384 tokens after resizing toward roughly 800 by 800 pixels, and bills at ordinary V4 Flash rates. Good for screenshots, charts, and document layout; not for small type or dense diagrams.
Multimodal model and API
Meta Muse Spark →
Meta's natively multimodal reasoning model, updated to version 1.3 on September 2, 2026. Reads images, charts, and text together, and offers a Contemplating mode in which multiple agents reason in parallel before answering. Hosted and proprietary at $1.25 per million input tokens and $4.25 output; Meta says an open-weights release is on the roadmap but has not given a date.
Music generation API
MiniMax Music 3.0 →
Production music model that takes a creative concept and optional lyrics and composes, arranges, performs and produces a complete song in a single generation, with instrumental-only support. Callable through MiniMax's platform API as model music-3.0, with open weights also published.
Music generation model
MiniMax Music 3 →
Open-weight model that generates complete five-minute songs with vocals in 32 kHz stereo from lyrics plus a structured style description. Runs via SGLang-Omni, Diffusers, or ComfyUI. Commercial use allowed with on-screen attribution; written permission required above $20M revenue.
News filter
unslop.news →
A Hacker News front page with stories about AI filtered out, built after the week's Ask HN complaint about the AI news flood. It uses a language model to classify each story's main subject; the code is open source.
Number-theory toolkit
CADO-NFS →
The open-source number field sieve implementation that Cognition modified for GPUs to factor RSA-260. The canonical starting point for anyone doing serious integer factorisation work.
Office-document access for agents
OfficeCLI →
A command-line tool that lets AI agents read and edit Word, Excel, and PowerPoint files, one of the week's fastest-rising agent-infrastructure repos on GitHub.
On-device AI
Program-as-Weights demo →
A public demo and code for compiling natural-language task specs into tiny neural artifacts that run locally on a frozen small model, matching much larger models on narrow fuzzy tasks.
On-device AI SDK
Desert Ant Core SDK →
SDKs for Swift, Kotlin and JavaScript that run 18 small models fully offline on iOS, macOS, Android and the web via Core ML, LiteRT and WebAssembly. Covers transcription, PII redaction, language detection, moderation, summarisation and video clipping with no API key and nothing leaving the device.
On-device LLM runtime
Cactus Hybrid (Gemma-4 E2B) →
A phone-sized Gemma-4 checkpoint with an attached error probe that scores how likely each answer is wrong and routes low-confidence queries to a cloud model; weights and runtime are public (set CACTUS_CLOUD_STRICT_SSL before using the cloud path).
On-device automation
Cactus Needle 3 →
Small sliceable models for structured extraction, embeddings and constrained function calling on devices.
On-device language model
Bonsai 27B →
PrismML's 1-bit and 1.58-bit builds of Qwen3.6 27B, compressing a 54 GB model to 3.9 GB (binary) or 5.9 GB (ternary) and running at roughly 11 tokens per second on an iPhone 17 Pro. The release ships an honest benchmark table showing the cost: instruction following, tool calling, and vision all degrade sharply, and the vendor states agentic coding is not a strong target of this release.
On-device model
MiniCPM5-1B →
OpenBMB's dense 1B local model with Think and No-Think modes, trained with SFT, RL, and on-policy distillation. Designed for on-device and edge deployment.
On-device model (preview)
Edge0-35B-A3B-preview →
A preview mixture-of-experts model released 8 September under Apache 2.0, with a runtime that streams experts from the SSD so the model can run in a small memory footprint on a laptop. The weights are a 19.7 GB download; speed figures on the card are the developers' own.
On-device model compiler
Program-as-Weights →
Turns a plain-English task spec into a small weight file that a frozen 0.6B model runs locally -- matching a 32B model's quality at roughly one-fiftieth the memory and about 30 tokens/sec on a MacBook M3. Open repo and site for compiling cheap, offline 'fuzzy' text programs.
On-device model runtime
Noema Overfit →
Repackages compatible mixture-of-experts model files so shared weights stay resident in memory while expert weights stream from local storage on demand, letting phones load models far larger than their RAM. Experimental, and slower than a smaller fully-resident model on short prompts.
On-device quantized model
Ternary-Bonsai-27B (GGUF) →
PrismML's ternary-weight 27B model in GGUF at ~7.2 GB deployed, with custom CUDA/Metal/CPU kernels. Expands local hardware reach, though agentic reliability is still limited per early tests.
On-device speech recognition
Apple SpeechAnalyzer →
Apple's on-device speech-to-text API that cut errors roughly fourfold over the legacy recognizer and beat Whisper Small using about a third of the compute - private, local transcription with no cloud round-trip.
On-device speech transcription
Voz →
Desert Ant's headline transcription model, a 489 MB Core ML package that the company says handles ten minutes of audio in about two seconds on a recent iPhone. Built on NVIDIA's Parakeet and optimised for Apple's Neural Engine.
On-device world model
Cosmos3-Edge →
NVIDIA's compact 4-billion-parameter physical-AI model generates text autoregressively while producing image, video, audio and action-trajectory outputs through a diffusion tower, sized for local robotics, autonomous-vehicle and smart-infrastructure workloads. NVIDIA warns it is not physically accurate simulation or safety-certified reasoning.
Open LLM
Qwen3.6 (open weights) →
Alibaba's stable Qwen3.6 release: open-weight general chat and coding models you can self-host, the same family at the center of this week's open-vs-closed pricing debate.
Open LLM weights
Xiaomi MiMo-V2.5-DFlash →
Xiaomi's official DFlash release on Hugging Face -- a 1-trillion-parameter mixture-of-experts model (42B active) under an MIT license, with FP4 quantization and parallel decoding for high inference throughput.
Open coding model
LongCat-2.0 →
Meituan's 1.6T-parameter MoE model tuned for coding and agentic work, MIT-licensed weights plus a cheap hosted API (launch promo $0.30/$1.20 per million tokens) that self-hosts to avoid data-jurisdiction concerns.
Open image model
Qwen-Image-2.0-Pro →
Alibaba's latest open image-generation model in the Qwen family, downloadable and runnable locally, part of a broad open-weight release wave that also refreshed the Qwen3.6 chat models.
Open language model
LLaDA / iLLaDA →
An openly released diffusion language model (weights and code) that generates text by refining a whole passage at once rather than one word at a time, useful for experimenting with non-autoregressive generation and infilling.
Open large language model
GLM-5.2 →
A flagship openly-available language model with a very large context window for long documents and code. Free to download and run yourself, with compressed versions for more modest hardware.
Open model
GLM 5.2 (GGUF, runnable locally) →
Zhipu AI's open, MIT-licensed mixture-of-experts model with a roughly million-token context, now packaged as ready-to-run quantized files you can host on your own machine. Strong on agent and coding workflows; this week it beat Claude on a narrow security benchmark at a fraction of the cost.
MiMo-V2.6-Distill-Qwen-9B →
Xiaomi's 9B Qwen3.5-based supervised distill for coding, agent, visual-coding, and cybersecurity research.
Open model download
Kimi K2.6 weights (Hugging Face) →
The actual Kimi K2.6 model weights, published under a modified-MIT license for anyone to download, run, and build on; large enough that full-strength use needs a multi-GPU node.
Open model for formal math
Leanstral 1.5 →
A free, open mixture-of-experts model specialized for writing machine-checked Lean 4 proofs and translating ordinary math into formal, verifiable form.
Open model weights
Frontis-MA1-35B →
A 35-billion-parameter open model post-trained specifically to write, run, debug and recombine machine-learning code inside an evolutionary search loop. Released with the full OpenMLE stack, so the search framework it was trained for is public too.
Open robotics model
RxBrain (Hy-Embodied-RxBrain-1.0) →
Tencent's ~6.2B embodied model that interleaves text reasoning with generated goal images to plan robot tasks. Weights and inference code released under Apache-2.0.
Open video understanding model
VideoChat3-4B →
A fully open 4B-parameter video multimodal model for general, long-form, and streaming video understanding, released with weights, training code, training strategy, and datasets.
Open-source agent framework
OpenClaw →
The fastest-growing repo on GitHub, now a MIT-licensed nonprofit, a neutral open framework for building AI agents that plug into any model or lab.
Open-source agent runtime
Codex Harness →
The Apache 2.0 agent runtime behind OpenAI's Codex, usable directly if you would rather run the loop yourself than buy the managed version. Ships as a CLI for bounded jobs, an SDK for programmatic control, and an app-server for persistent threads with human approvals.
Open-source speech recognition
OpenAI Whisper →
OpenAI's open-source speech-recognition model family and the reference baseline Apple's SpeechAnalyzer was measured against - freely runnable locally in sizes from tiny to large for transcription and translation.
Open-weight LLM
Qwen3-Next-80B-A3B-Instruct →
Alibaba's efficiency-focused open-weight model (80B total / 3B active, 512 experts) with 262K native context to ~1M, built around hybrid attention, high-sparsity MoE, and multi-token prediction; the model card claims roughly 10x inference throughput past 32K context versus a dense 32B baseline.
Qwen3.6-35B-A3B →
Alibaba's open-weight agentic-coding model (35B total / 3B active, Apache 2.0) with 262K native context extensible toward 1M tokens, hybrid Gated-DeltaNet + MoE attention, thinking preservation across turns, and built-in tool use. Downloadable and self-hostable on common open serving stacks.
Open-weight agent model
Solar Open 2 →
Upstage's 250-billion-parameter mixture-of-experts model activates only 15 billion parameters per token and runs on two NVIDIA H200 GPUs once quantized, with a one-million-token context aimed at long multi-step agent work. Weights and a full technical report are public under a custom license requiring Solar-prefixed derivative names and Built with Solar attribution.
Open-weight coding model
Kimi K2.7 Code →
Moonshot AI's trillion-parameter mixture-of-experts coding agent, with only 32B active per token, a 256K context, and vision input, now selectable inside GitHub Copilot and downloadable under a Modified MIT license.
Poolside Laguna S 2.1 (GGUF) →
Open-weight 118B mixture-of-experts coding agent activating about 8B parameters per token, under the permissive OpenMDW-1.1 licence, in GGUF plus FP8, NVFP4 and INT4 builds. Use the current re-released Q4/Q8 files - the initial ones shipped with a broken chat template.
Open-weight language model
Gemma 4 26B A4B →
Google's compute-efficient multimodal model with 25.2 billion total parameters but only 3.8 billion active per token, aimed at running usefully on hardware that cannot host a dense model of comparable capability.
Ling-3.0 →
inclusionAI's hybrid-linear mixture-of-experts family under a plain MIT license, mixing three linear-attention blocks per full-attention block across 128 routed experts. The tiny variant holds 7.9B parameters and activates 1.3B per token; native BF16, FP8 and INT4 support is declared on the card.
Ornith 1.0 →
An MIT-licensed, Qwen 3.5-derived family published at 9B, 35B and 397B on Hugging Face. Worth pairing with the model's public discussion threads before deploying, where users have been diagnosing apparently missing multi-token-prediction tensors in the shipped checkpoints.
Open-weight language models
DeepSeek-V4 (Pro & Flash) →
Two newly previewed open-weight models with a 1-million-token context window on by default - a large mixture-of-experts flagship and a smaller, fast everyday model. Downloadable weights plus an API.
Google Gemma (open weights) →
Google's open-weight model family, light enough that developers are now embedding it directly into interactive apps - including a demo running Gemma inside the Godot game engine via Vulkan compute shaders, no Python server required.
Open-weight local model
Qwen3.8-27B →
Apache 2.0 vision-capable 27B model with a 262k context window, runnable on a well-specced laptop in quantized form. Ships with reasoning effort set to xhigh, which is worth turning down before first use.
Open-weight model
DeepSeek V4 Flash 0731 →
The current V4 Flash checkpoint, with weights, the DSpark draft head embedded, and the encoder file that reveals the reasoning-effort labels are prompt prefixes rather than a compute dial. The card also specifies the intended FP8 key-value cache and FP4 indexer cache serving recipe.
DeepSeek-V4-Flash →
MIT-licensed weights for DeepSeek's 284B-total / 13B-active mixture-of-experts model with a one-million-token context, with vLLM and SGLang serving examples on the model card. Real hardware bar: the reference recipe targets four B200 or B300 GPUs.
DeepSeek-V4-Pro →
A downloadable 1.6-trillion-parameter mixture-of-experts model that activates 49 billion parameters per token, with a one-million-token context window under an MIT license. Serious server hardware required, but the weights are yours.
Ling-3.0-flash →
inclusionAI's 124B mixture-of-experts model with about 5.1B parameters activated per token. Sparse routing genuinely cuts per-token compute, but this is a server-class artifact, not a laptop one: the BF16 repository is roughly 255GB and the official serving path calls for custom SGLang or vLLM forks with tensor parallelism across four GPUs.
MiniMax-M3 →
A natively multimodal open model trained on text, image, and video from the first step, with a million-token context and a sparse-attention design built for speed; downloadable for self-hosting and also offered through MiniMax's own API and agent platform.
Muse Glimmer 30B →
Meta's 30-billion-parameter open-weight agent model under Apache 2.0, built for always-on local workflows with text and image input, tool use, a context window past 131,000 tokens, and a speculative decoder that drafts sixteen words at a time. Full weights, quantized builds, and the drafter are all in the release.
Ornith-1.5-35B-A3B →
The tool-using, agentic-coding mixture-of-experts base model behind TielCoder, with long context and a vision tower. Its multi-token-prediction head was re-uploaded in trained form on August 23, 2026.
Qwen3.8-Flash-Next →
Alibaba's preview of the architecture behind Qwen4: 125 billion parameters with 6 billion active, a 20-million-entry n-gram embedding table, and a 262k context extensible to a million tokens. Weights are 360 GB in bf16 under the Qwen Community License 1.0, which allows commercial use and fine-tuning but requires a separate licence to run a model-as-a-service or an AI coding-assistant business.
Tencent Hy4-preview →
Tencent's Apache-2.0 licensed mixture-of-experts model, 770B total parameters with 49B active and a context window over one million tokens, aimed at agentic coding and long-horizon tasks. The full repository is about 1.56 TB on disk, so this is server hardware territory, not a local model.
Open-weight model (agent world model)
Qwen-AgentWorld →
Alibaba's open language world model that simulates agent environments -- browser, terminal, phone, coding workspace and more -- so other agents can be trained inside the simulation. Released with open weights and code in two sizes.
Open-weight model (self-host)
DiffusionGemma →
Google's open-weight text-diffusion model that generates text in parallel blocks instead of one token at a time; Apache-2.0, runnable locally, with community tooling already shipping.
Open-weight model family
K2 Horizon (IFM) →
Six Apache 2.0 models from 375B-A23B down to 0.9B, sharing architecture, vocabulary and training methodology, with 512K context on all but the smallest. The 36B-A4B is a 74.9 GB bf16 download or 48.4 GB in FP8, with a serving recipe validated on two H200 GPUs. Upstream llama.cpp support is still in progress.
Ornith-1.5 →
Three open coding and agentic models -- 397B and 35B mixture-of-experts plus a 9B dense model with a quantized Mobile build for phones. The 9B is single-GPU at roughly 19 GB with a 262,144-token context and OpenAI-compatible tool calling; the flagship matches Claude Opus 4.8 on terminal-coding benchmarks.
Open-weight multimodal model
GLM-5.3-Flash →
Z.ai's 320-billion-parameter multimodal model with 18 billion active per token and a one-million-token context, released under the MIT licence -- one of the most permissive terms any model this size has shipped under. The download is 328 GB of already fp8-quantized weights, and it runs locally through SGLang, vLLM or TokenSpeed.
Open-weight multimodal models
Gemma 4 →
Google's downloadable model family (2.3B-31B, dense and MoE) that natively handles text, vision, and audio, including a 12B encoder-free variant and a thinking mode.
Open-weight reasoning model
Ring-2.6-1T →
Ant Group's trillion-parameter mixture-of-experts reasoning model, activating roughly 63 billion parameters per token, with 128K context extendable to 256K. All checkpoints openly downloadable under the MIT license, with high and xhigh reasoning-effort settings that trade depth against speed and cost. Benchmark claims are vendor-supplied and measured against a previous generation of rivals.
Open-weight research agent
AREX-Turbo →
Apache-2.0 4-billion-parameter deep-research agent from BAAI that audits its own provisional answers against the question's constraints and re-runs research when confidence is low. The public quick-start exposes search and page-visit tools but not the paper's full outer control loop.
Open-weights agent model
Nex-N2.5-Pro →
Released 8 September under Apache 2.0: a 397-billion-parameter model from Nex AGI aimed at agentic work such as computer use and browsing. The repository's files total 407 GB.
Nex-N2.5-mini →
The smaller sibling of Nex-N2.5-Pro, released 8 September under Apache 2.0: 35 billion parameters, a 70.2 GB download.
Open-weights math model
Nemotron IMO 2026 checkpoints →
NVIDIA's released specialist checkpoints, training data, 200-problem benchmark and inference recipe behind an officially graded IMO 2026 gold-level score (30 of 42). Each checkpoint is a 1.12 TB download; the model card recommends at least eight B200 GPUs.
Open-weights model
DeepSeek-V4.1-Flash →
DeepSeek's new MIT-licensed model, released 10 September: a 552B backbone plus a 196B lookup-based memory module, with native image input and a one-million-token context. The published weights total 510.3 GB across 48 shards.
Inkling →
Thinking Machines Lab's 975B-parameter mixture-of-experts model, released July 15 under Apache 2.0. Only ~41B parameters activate per token, it accepts text, image and audio input, and it handles up to 1M tokens of context. Artificial Analysis ranks it the top US open-weights model. Free to download, modify and use commercially -- but you will need serious hardware to run it.
Open-weights multimodal model
Agnes-3.0-Flash (Preview) →
Agnes AI's 33-billion-parameter text, image and video model, released 11 September under Apache 2.0, with a 262,144-token context and a hybrid recurrent-plus-attention design. A 66.2 GB bf16 download; the model card asks for one NVIDIA H200 (141 GB) or H100 (80 GB) plus 128 GB of host memory, and a community Q4_K_M GGUF is 19.75 GB. It is a Preview checkpoint: the Artificial Analysis score belongs to a different API model.
Open-weights music model
YuE2-3B →
M-A-P's song generator, released 9 September: writes an editable melody-and-chord score in ABC notation, then renders the full song. A 7.3 GB download plus a 0.5 GB decoder; the model card asks for a 24 GB NVIDIA GPU. Weights are CC BY-NC 4.0 (non-commercial), code Apache 2.0.
Open-weights science model
Intern-S2-397B (FP8) →
The InternLM team's multimodal model for scientific reasoning and long agent tasks, a mixture-of-experts design with 512 experts and 10 active per token, released under Apache 2.0. The FP8 version is a 406.3 GB download (the bf16 repo is 806.9 GB), and InternLM's deployment guide recommends a node of eight H100 or eight H200 GPUs. Serves with LMDeploy, vLLM or SGLang.
Open-weights terminal agent model
T1-122B-A10B →
A 122-billion-parameter command-line agent model from a Tencent-led team, trained with reinforcement learning on 15,000 synthetic shell tasks and released under CC BY 4.0 (the Qwen3.5 base keeps its Apache 2.0 licence). It is a 250.2 GB download in bf16; the model card states no GPU requirement, and the weights alone occupy about 250 GB before any cache. The paper reports it matching Claude Opus 4.6 on a terminal-task benchmark while taking about three times as many turns as its starting checkpoint.
Personal AI agent
Meta Muse →
Meta's consumer agent, launched 8 September in the US on iOS, Android and muse.ai, runs each user's tasks on a dedicated cloud computer with its own browser, and a separate 'Sentinel' agent must approve anything that reaches the internet. Users choose whether it can only read email or also send it. Free for most use, with paid plans reported at $20 and $100 a month; interactions train Meta's models unless you opt out.
Plugin framework
Cordis →
The plugin framework underneath DeepSeek Harness, describing itself as a meta-framework for spatiotemporal composability. It supplies runtime mount and unmount of components with revertible effects and reactive dependencies, so parts of a running system can be swapped while the pieces depending on them react correctly instead of silently breaking.
Post-training toolkit
SLAI T-Rex →
The public workflow behind a full-parameter Ascend post-training run on a DeepSeek-V4-family model: FP8-to-BF16-to-Megatron checkpoint conversion, launch templates, and inspectable data-construction pipelines for continued pre-training and supervised fine-tuning. The production engine and custom kernels are withheld.
Pretraining dataset
Nemotron-ClimbMix →
NVIDIA's curated pretraining corpus, and the data behind this week's $998 from-scratch training run. A practical way to substitute somebody else's filtering work for compute you would otherwise spend on mediocre tokens.
Pricing reference
Anthropic prompt caching pricing reference →
Anthropic's documentation of how cached tokens are billed: cache hits at 10% of standard input, five-minute writes at 1.25 times base input, one-hour writes at 2 times. This is the page that explains why Claude Fable 5.1 can be substantially cheaper per prompt for long agent sessions while costing exactly the same for one-shot calls.
Gemini Developer API pricing page →
Google's own current rate card, including the context-caching rates that determine whether a long-running agent is cheap or ruinous. Worth reading before assuming a headline per-token price describes what you will actually pay, since caching rates and introductory-period expiry dates do most of the work.
Proof tooling
Lean comparator →
The Lean toolchain component used to independently check the formalization behind Anthropic's zeta-function result. Useful to anyone who wants machine-verified mathematics rather than a persuasive argument.
Protein and antibody design
Chai Discovery design suite →
An interactive design canvas for biologics rather than a prediction tool: choose an antibody format, choose or infer the target structure, pick the epitope, target a specific antigen state, and specify chemical modifications. Commercial access with limited academic availability.
Protein testing service
Adaptyv Bio →
A contract lab that will express and measure binding for protein designs you submit, including designs produced by a model. One of the two independent labs that validated Anthropic's Claude-designed binders.
Provenance standard and tooling
C2PA Content Credentials →
Open standard for cryptographically signed provenance metadata in media files, now used by Claude to tag generated images and by camera makers and photo editors to record where a file came from. Any C2PA-aware tool can read the credential.
Public data API
Stack Exchange API →
The free, key-less API behind Stack Overflow and its sister sites, which will return exact question, answer and user counts for any date range -- useful for checking claims about the site's decline yourself.
Public research leaderboard
Elliptic Curve Rank Leaderboard →
An NSF-funded public record of high-rank elliptic curves, where every submission publishes its witness points, commentary and edit history, and offers a JSON endpoint so anyone can verify a claimed record independently.
Python library
AlphaGenome API client →
Google DeepMind's open-source client for querying AlphaGenome variant-effect predictions from Python, for researchers who want programmatic access rather than the web portal.
Quantised model weights
Unsloth DeepSeek-V4-Flash-0731 GGUF →
Published quantisations of DeepSeek's 671-billion-parameter Flash model, ranging from roughly 91 GB at two bits to 162 GB at eight. The card is also the clearest available statement of what hardware each tier actually needs.
Quantization toolkit
TurboQuant-MLX →
Quantization tooling for MLX with published size and speed measurements, including a 3-bit path that takes a 120-billion-parameter model from about 63GB down to 48GB on consumer Macs.
Quantized model build
Qwen3.8-Flash-Next GGUF (Unsloth) →
Unsloth's quantized GGUF conversions of Qwen3.8-Flash-Next, including a 2-bit UD-Q2_K_XL build at roughly 78.9 GB across three shards -- about a fifth of the official bf16 repository. The model card carries working setup instructions for llama.cpp, vLLM and Ollama.
Quantized model builds
Muse Glimmer 30B GGUF (Unsloth) →
Community-packaged quantized builds of Muse Glimmer that fit under 20 GB, with setup instructions for llama.cpp, Ollama, vLLM, and SGLang. This is the practical path if you want the model running on a single 24 GB consumer graphics card rather than compiling the full-precision weights yourself.
Quantized model files
Qwen3.8-27B GGUF (bartowski, new tensor layout) →
Re-uploaded 11 September using per-tensor quantization recipes, which the author measured at lower KL divergence than the standard layout at the same file size (Q4_K_M at 0.94x). The Q4_K_M file is now a 17.44 GB download; the recipes ship in a layouts folder for reuse with llama.cpp.
RL training framework
slime →
The open-source large-scale asynchronous training framework from THUDM that Z.ai used to run the post-training scaling behind GLM-5.3.
Real-time interactive video
Wan Streamer v0.3 →
Streaming audio-visual interaction model that treats a video as a persistent world plus a time-varying event stream, running full-duplex real-time conversation at 640x368 / 25fps with roughly 550ms total interaction latency.
Real-time voice
GPT-Live →
OpenAI's full-duplex voice interface that talks, listens, and interrupts in real time while delegating deep reasoning to GPT-5.5 in the background; free mini tier plus a paid tier.
Realtime voice API
Gemini 3.8 Live →
Google’s low-latency multimodal voice endpoint supports streaming audio, visual input and asynchronous function calls.
Reasoning assistant + API
Muse Spark 1.1 →
Meta Superintelligence Labs' multimodal reasoning model built for agentic work - tool and computer use, coding, a 1M-token context window, and subagent orchestration; live in the Meta AI app's Thinking mode and on meta.ai, with a Meta Model API in public preview.
Reasoning voice API
Gemini 3.8 Live Extended Thinking →
A separate Gemini Live endpoint for multi-step spoken tasks with background reasoning, progress narration and asynchronous tools.
Regulatory compliance reference
AI Act Service Desk →
The European Commission's official article-by-article guide to the AI Act, including Article 50 transparency duties that apply from 2 August 2026. Note its own warning that displayed text may lag the latest amendments - check the Official Journal for dates.
Reinforcement-learning training framework
prime-rl →
Open-source RL post-training stack that splits rollout generation and gradient updates across GPUs, used in this week's widely discussed $500 fine-tune that beat five frontier configurations on a catalog-review workflow. Practical for teams that already have an automatically scored task and want to train a specialist rather than pay per call for a frontier model.
Reliability testing
SC25 LLM reliability assessment →
Fault-injection harness that flips individual bits during language model inference through PyTorch hooks, then restores them, so you can measure how your own model degrades under simulated soft errors instead of assuming it is resilient.
Research agent framework
OpenResearch →
alphaXiv's open-source tool for running parallel research agents on any model; version 0.2.0 shipped 11 September under the MIT licence.
Research dashboard
AI Data-Center Tracker →
A live map of the US AI data-center buildout covering 1,547 facilities across 46 states, 83 frontier sites with named megawatt figures, and 530 state and federal bills, refreshed hourly from public feeds with per-figure source and confidence labels.
Research model checkpoints
NCP-ArchPreview (8.9B) →
Shanghai AI Lab and SJTU's language model that predicts compressed 'concepts' alongside the next token, reaching OLMo-3-7B's final pretraining loss with about half the training tokens. Apache 2.0; each 8.9B checkpoint is a 17.9 GB download.
Research toolchain
transformer-vm →
Compiles C programs to WebAssembly and then into analytically constructed transformer weights, with a C++ engine that executes them inside the model at about 30,000 tokens per second.
Research verification
SAI ICML 2026 replication results →
A browsable record of automated replication attempts against all 168 oral papers from ICML 2026, showing which papers shipped runnable code and how many of each paper's claims actually reproduced. Useful before you build on a result you have only read the abstract of.
Research workbench
Claude Science →
An AI workbench that unifies literature search, notebooks, statistics, and cluster compute, and keeps a reproducible record behind every figure. Beta on Mac and Linux.
Retrieval library
PyLate →
A training and retrieval library for late-interaction models, built on Sentence Transformers, for people who want to fine-tune a retriever on their own corpus rather than use an off-the-shelf embedding API.
Retrieval model
DenseOn →
A fully open 149-million-parameter dense retrieval model from LightOn for multilingual, long-context and code search, released with its training data and training code rather than weights alone.
LateOn →
LightOn's late-interaction counterpart to DenseOn - it keeps a vector per token instead of one per document, which costs more storage but retrieves noticeably better on hard queries.
Reward model
OS-Shepherd-9B →
A 9B reward model trained specifically to judge whether a computer-use agent actually finished its task, built to cut the false-success verdicts that general-purpose vision-language judges produce. A 35B sibling and the OSReward benchmark ship alongside it.
Robot manipulation dataset
HiFi-UMI-2K →
Released dataset behind this week's handheld-only robot training result: high-fidelity two-handed human demonstrations captured with a head-mounted stereo rig and tracked grippers, with every trajectory reconstructed and rejected unless a target robot could physically replay it. Covers wiping, shirt folding, remote insertion and produce sorting. Directly usable for imitation-learning experiments without owning a teleoperation setup.
Robot navigation model
Robostral Navigate →
Mistral's 8B embodied model that steers wheeled, legged, or flying robots through unseen environments from a single RGB camera and a plain-language instruction.
Robot policy model
WorldDiT →
Four released checkpoints plus self-contained inference and evaluation code for a sub-billion-parameter diffusion transformer that emits continuous robot action chunks while predicting future camera-frame pixels as auxiliary training signal. The visual prediction head is dropped at deployment. Tested across four LIBERO simulation suites; the model card notes its cross-paper comparison mixes published protocols.
Robotics API
Gemini Robotics ER 2 →
The embodied-reasoning half of Google DeepMind's new robotics family, and the only part available now - it reasons about physical scenes and plans robot tasks via the Gemini API and AI Studio, while the models that actually drive motors stay in private preview.
Run AI models efficiently
SGLang v0.5.13 →
A high-performance open serving engine for language models. The new version turns on faster 'guess-ahead' decoding by default and trims scheduling overhead for quicker responses.
vLLM v0.23.0 →
The widely-used open engine for serving language models fast and cheaply. The latest release adds smarter memory handling for long conversations and faster GPU execution.
Run models on your computer
LM Studio →
A friendly desktop app to find, download, and chat with open models on your own machine — no command line needed.
Ollama →
Download and run open AI models locally with a single command. The easiest on-ramp to running your own model.
Open WebUI →
A polished, ChatGPT-style web interface for the open models you run yourself.
llama.cpp →
The lean, fast engine that makes big models run on ordinary laptops; powers much of the local-AI ecosystem.
SSD-streaming inference engine
Mference →
Runs DeepSeek V4 Flash on Apple silicon by keeping the shared core, attention and cache resident while streaming each token's routed experts off the SSD. Publishes an unusually honest memory budget: about 3 GB working set, 90 to 98 GB on disk, tested at a 4,000-token context on a 24 GB Mac, with no quality parity test yet.
Safety classifier
Shieldstral 1.0 3B →
Mistral's open-weight multimodal moderation model. You supply the policy as a plain-language yes/no question at inference time rather than retraining for a fixed harm taxonomy, and it returns one calibrated safety score per forward pass. Handles prompts, responses, prompt-response pairs, images and image-plus-text across twelve languages. Apache 2.0, runs on a single 16GB GPU via vLLM, Transformers or llama.cpp; recommended operating context is 32k tokens.
Safety guardrail model
Qwen3Guard →
Alibaba's first open-weights safety-filter model, released under Apache 2.0 in three sizes, covering 119 languages, with a streaming variant that can flag unsafe text token by token as it is generated.
Scientific ML library
neuraloperator →
The open-source PyTorch library for Fourier neural operators and related architectures -- the toolkit behind FourCastNet and the plasma and lithography surrogates. If you want to try learning a solution operator for a PDE instead of solving it step by step, this is the reference implementation.
Scientific literature search
AskChem →
A live chemistry search system that retrieves individual claims rather than papers - 2.4 million typed claims from 147,000 papers, each carrying a source identifier and a verbatim quote. Free web interface plus REST, SDK and MCP access so agents can query the same claim store.
Scientific search engine
Large Discovery Models →
Released code for a discovery loop that pairs a generative proposer with a Bayesian surrogate scored on real experimental results rather than model confidence, applied to molecules, antibodies and training configurations.
Search agent
ABSeeker →
A released 4-billion-parameter web-research agent trained with per-step credit assignment that matches roughly 30-billion-parameter agents on hard fact-finding tasks.
Security AI platform
CrowdStrike SafeMind →
A paired offensive model (Red Tempest) and defensive model (Blue Solano) built on NVIDIA Nemotron and run inside harnesses that pit them against each other. Operates natively in the Falcon platform; standalone model access is gated behind the Project QuiltWorks program.
Security coding assistant
OpenAI Codex Security (Daybreak) →
An in-IDE plugin from OpenAI's Daybreak initiative that finds, validates, and fixes software vulnerabilities, plus an open-source remediation program run with Trail of Bits and HackerOne.
Security release
Datasette 1.0a39 and 0.65.4 →
Security updates for the open-source data publishing tool, released 10 September after an audit run with Claude Fable 5.1, GPT-5.6 Sol and GPT-6 Astra. Upgrade now if you run a public Datasette instance that gives authenticated users access to private data.
Security scanning
Claude Security →
Anthropic's code-security product for Enterprise plans, now running on Claude Mythos 5. An organization owner enables it in the admin console, and it follows a scan, validate, review, patch workflow, returning findings with weakness classifications, confidence and severity ratings, and suggested fixes rather than exposing the underlying model directly.
Self-developing coding agent harness
Ouroboros →
An agent harness that improves its own tools, prompts and core implementation through reviewed commits, which then become the runtime for its next task. Public code, with benchmark campaigns run on frozen snapshots so the numbers mean something.
Self-hosted AI companion
AIRI →
Self-hosted embodied assistant with a Live2D or VRM character, voice, persistent memory, local inference support, and game and chat integrations. A vertical application rather than a general agent framework.
Self-modifying agent runtime
EXO / exoharness →
Agent runtime that separates a disposable executor holding prompts, model calls and tool use from a durable harness holding the event log, secrets, artifacts and snapshots, so an agent can rewrite its own policy and still stop, resume, fork or rewind against intact history.
Serve at scale
vLLM →
The popular open engine for serving AI models fast and efficiently when you need to handle real traffic.
Serving recipe
SGLang K2 Horizon Cookbook →
IFM's validated serving configuration for the K2 Horizon family, with measured H200 latency and throughput for every model size. Covers the tensor-parallel setup and the router numerics override that preserves checkpoint behaviour -- the difference between the model running and the model running correctly.
Single-GPU inference engine
NInfer →
A focused inference engine that runs Qwen 3.6 models on one RTX 5090 with a 262,000-token context using an INT8 key-value cache, reporting roughly 188 tokens per second at 250,000 tokens of context. Methodology, seeds and limits are published openly.
Skills library for coding agents
agent-skills →
Addy Osmani's collection of production-grade, reusable skills for AI coding agents, trending near the top of GitHub this week.
Small language model
LFM2.5-2.6B →
Liquid AI's 2.7B tool-calling model with a 128k context, built as 22 short-convolution layers plus 8 grouped-query-attention layers so most token mixing stays local and cache-friendly. Post-trained inside real agent harnesses for tool use, extraction, retrieval and long-context workflows. The model card explicitly recommends against agentic coding and knowledge-heavy tasks, and it always enters a reasoning mode before answering.
Small looped-transformer model
Nanbeige4.2-3B →
An Apache-2.0 4B model that reuses one 22-layer transformer stack twice for 44 layers of depth from a single set of weights, shipping BF16 weights with SGLang, vLLM, llama.cpp, and Ollama paths for local use.
Small open models
Ternary Bonsai models →
A family of 1.7B, 4B and 8B models built for extreme quantization, shipped in the official group-64 two-bit format that mainline llama.cpp reads. Useful if you want to see what 2-bit inference feels like without converting anything yourself.
Speculative decoding draft model
Kimi-K3-DSpark →
Inferact's draft model for Kimi K3. It proposes seven tokens at a time for K3 to verify and accept or discard, and its block-diffusion backbone shares K3's attention-cache layout so no second cache format is needed. This is the component behind the 21-25 tokens per second measured on a sixteen-node GB10 cluster running the full K3 checkpoint.
Speech generation and editing
AuK →
Tencent's 1.5B open-weights speech model, MIT-licensed, that clones voices, edits recordings, cleans up audio and separates speakers through a single plain-language instruction interface. About 6.8 GB to download, with a four-step distilled variant for faster generation and a hosted demo to try first.
Speech model
NVIDIA NemotronLabs VoiceChat 11B →
An open-weight end-to-end full-duplex voice model that listens and speaks simultaneously and calls tools mid-conversation, shipped with both offline inference code and a containerised WebSocket streaming deployment. Needs an NVIDIA GPU with at least 80 GB of memory, and uses a single fixed voice.
Speech-to-text
Cohere Transcribe Arabic →
Open-source (Apache 2.0) Arabic speech-recognition model built for dialects and Arabic-English code-switching, with lower word error rate than Whisper Large V3 on the Hugging Face Arabic leaderboard.
Speech-to-text API
Gemini 3.5 Transcribe →
Google's new transcription model, shipped as two endpoints: a bidirectional streaming version for live voice agents and a batch version with speaker attribution and word-level timestamps. Handles 85+ languages with mid-stream language switching, cleans filler and self-corrections automatically, and can make function calls to other Gemini models. Try it in AI Studio.
Stealth browser / scraping engine
Fortress →
Open-core stealth Chromium with C++-level fingerprint patches that lets browser agents and scrapers pass bot detection (Cloudflare, DataDome, Turnstile); ships 29 pre-built MCP tools for the agentic web.
Stealth reasoning model
Ox Alpha →
An anonymous reasoning model with a 1,048,576-token context window and 131,072 max output tokens, accepting text, image, and video with tool and JSON support. Free to try in the browser, with no disclosed creator.
Structured decision API
TypeSafe Jev on Vercel AI Gateway →
A hosted typed-decision model exposed through Vercel AI Gateway and AI SDK 7's experimental evaluate API.
Terminal coding agent
Kimi CLI →
Moonshot's Apache-2.0 terminal agent for driving Kimi models from the command line for coding and tool use. Open-source software (distinct from the K3 model weights, due July 27).
Terminal editor
rune →
A terminal editor built for working alongside coding agents -- markdown-first with syntax highlighting, Obsidian-style vaults and wikilinks, tables, task lists, auto-merge and crash recovery. Its author rewrote all 65,000 lines from Go to Rust for about $400 using Claude Fable.
Text- and image-to-video API
fal H3 Max →
A post-trained MiniMax H3 that renders a five-second 768p clip in under three seconds through fal's API, at 480p or 768p and up to 15 seconds. Hosted only -- there are no downloadable weights -- and priced at $3.60 per minute of generated video. There is a browser sandbox for trying it without writing code.
Text-to-speech model
Inflect-Micro-v2 →
A complete English speech synthesis stack in 9,356,513 parameters, waveform decoder included, producing 24 kHz mono audio locally with no external vocoder or API. One fixed synthetic male voice, no cloning, flatter prosody than large systems - but it runs anywhere.
Tokenizer library
GigaToken →
An open-source native BPE tokenizer optimized with SIMD byte scanning and instruction-level parallelism, best used for fast offline corpus preparation and bulk token counting rather than end-to-end serving speedups.
Tool-calling evaluation
tool-eval-bench →
A public benchmark harness for testing tool calling against OpenAI-compatible local and hosted serving endpoints.
Train & fine-tune AI models
veRL →
The open RL post-training framework used by most research labs training reasoning models today. Run GRPO, PPO, and related reward-training methods on your own models.
Training data and checkpoints
FACET terminal-agent task set →
A public release of 6,020 synthesized terminal-agent tasks plus three fine-tuned checkpoints. Each task bundles an instruction, an initialized environment, a reference solution, and an executable verifier, all grounded in the same container state so they cannot drift apart. Directly usable as reinforcement-learning environments for coding and shell agents.
Training dataset
Recursive-Task-Synthesis →
A public set of 37,484 verified long-horizon terminal-agent tasks, each a runnable bundle with instruction, environment, reference solution and hidden verifier, plus a companion set of 327,000 agent trajectories and three fine-tuned Qwen3.5 checkpoints.
The Stack v3 →
Hugging Face's code corpus, now with source text embedded inline rather than behind identifiers. A 15.9 TB deduplicated, PII-redacted training split of roughly 4.9 trillion tokens, plus a 113.7 TB unfiltered bucket for teams that want to build their own mix.
Training framework
MindSpeed-LLM →
Huawei's official large-model training toolkit for Ascend NPUs, covering distributed layouts, checkpoint conversion and supported model families. Worth reading its support table honestly -- DeepSeekV4-Flash is currently marked Prototype, its label for not-fully-validated features.
Training method
SimpleOPD →
An on-policy distillation implementation that works between models with different tokenizers, aligning only the text spans both vocabularies agree on. That lets a teacher in one model family train a student in another without building a vocabulary bridge.
Training reference implementation
smol-kimi-k3 →
A runnable 49-million-parameter from-scratch trainer that deliberately keeps Kimi K3's unusual architectural ingredients rather than collapsing into a vanilla transformer, including the 3-to-1 KDA and gated MLA rhythm, ShortConv-based KDA, attention residuals, stable latent mixture of experts, and no positional encoding. Ships the recipe, a 10k-step checkpoint and verification checks, and is explicit that it is an architecture study rather than a capability claim.
Usage data
OpenRouter Rankings →
Public leaderboard of which models are actually receiving token traffic, broken out by task category and time window. The clearest free read on real-world model adoption rather than benchmark scores.
Vector search
FastPLAID →
A Rust engine for multi-vector search, the indexing layer that makes late-interaction retrieval fast enough to serve in production instead of only benchmarking well.
Video and audio generation
MiniMax-H3 →
Open weights for MiniMax's omni-modal model that generates four to fifteen second video with native stereo audio. The locally deployable base runs at 768p through diffusers or SGLang; the prompt-interpretation and 2K regeneration stages stay behind MiniMax's API, and the licence excludes the US, EU, UK, and South Korea.
Video evaluation toolkit
FilmOps + FilmBench →
Public benchmark assets for judging generated video on professional film craft instead of generic prettiness. FilmOps ships six specialized operators covering shot scale, composition, camera angle, color and tone, character layout and camera movement; the companion FilmBench dataset supplies prompts reverse-engineered from professionally selected clips, most of which require multi-shot continuity. Authors report weaker agreement with human raters on audio and editing than on visual categories.
Video generation
Seedance 2.5 on Dreamina →
ByteDance's newest joint audio-video model, announced 31 July, generating a single take of up to 30 seconds extendable twice, with white-model control, green-screen editing and camera and blocking controls. Note that the 4K output, 50-reference limit and 180-second beta advertised on this page are marked Coming Soon.
Vidu (Vidu S1 Stream Model) →
A working AI video generator with text-to-video, image-to-video, and reference-to-video modes; the new S1 Stream Model targets real-time, interactive, voice-steerable video at up to 42 FPS/540p on consumer GPUs. Free credits to try, paid plans for more.
Video generation API
Gemini Omni 1.1 Flash →
Google's production-ready generative video model, now able to extend an existing clip using up to ten seconds of prior context, generate between specified first and last frames, draft at 360p for about a third the cost, and upscale finals to 4K. API-only through Google AI Studio.
Gemini Omni Flash →
Google's new video model offering developers programmable conversational editing - generate and revise clips up to ten seconds by describing changes in words.
Video generation model
LTX-2.5 →
Lightricks' open-weight video foundation model, shipped 11 August with a new diffusion decoder, native multishot generation, 4K HDR and automatic clip-length prediction. Runs locally on 16GB of VRAM, or through a per-second API. Free for commercial use below $10M annual revenue.
Video generation plugin
10S-Comfy-nodes (LTX Tiled Sampler) →
A drop-in replacement for ComfyUI's SamplerCustomAdvanced that fixes LTX 2.5's broad colour smearing by sampling in spatial tiles, keeping each tile inside the model's training token count instead of running a 2x-upscaled latent, then stitching with cosine-Hann overlap blending. Actively maintained with external contributions.
ComfyUI-SolAttn-H3 →
A ComfyUI custom node that wires NVIDIA's training-free Sol-Attn sparse attention into the native MiniMax-H3 video path using public ModelPatcher APIs, so it drops in without modifying any ComfyUI core files. Keeps a prefix sink exact and sparsifies the remaining attention via a tunable threshold, with the gain growing as sequence length grows.
Video understanding model
Mage-VL →
Microsoft's codec-native multimodal model that reuses a video file's own bit allocation to pick visual tokens, reporting over 75% fewer tokens and up to 3.5x faster inference than uniform frame sampling. Works with H.264, HEVC and DCVC-RT.
Vision-capable model endpoint
DeepSeek V4-Flash vision API →
Experimental image input for DeepSeek's cheap V4-Flash model, billed at the same rate as the text-only version. Accepts images as inline base64, as a URL the model fetches, or as a file uploaded through the Files API. The experimental tag is real, so treat the interface as unstable, but it makes high-volume image reading economically sensible.
Vision-language model
moondream 3.1 (9B-A2B) →
An open-weight vision-language model with 9B total but only 2B active parameters, offering native object detection, pointing, captioning, and segmentation at roughly the speed of a 2B dense model.
Visual prototyping
Claude Design →
HTML-based Claude canvas for editable mockups, one-pagers, visual flows and presentation assets.
Voice agent
Realtime-Venus →
Public full-duplex audio and audio-visual agent models with interruption handling and asynchronous delegation demos.
Voice agent framework
LiveKit Agents →
Production framework for building realtime voice agents, with interchangeable speech-to-text, LLM, text-to-speech, and realtime components plus semantic turn detection. This is the plumbing layer around a voice model rather than a duplex model itself, and it trended on GitHub today.
Voice agents
speech-to-speech →
Hugging Face's modular local voice-agent pipeline - voice detection, speech recognition, a language model and text-to-speech chained together, with an OpenAI Realtime-compatible websocket so existing clients can point at it.
Voice pipeline
Hugging Face speech-to-speech →
Local voice-activity detection to speech recognition to language model to text-to-speech pipeline, threaded through queues and exposed as an OpenAI Realtime-compatible server so existing clients can point at it unchanged.
Weather data
WeatherNext 3 forecasts →
Google's hourly global AI-weather forecasts are requestable through its documented Google Cloud, BigQuery and Earth Engine access path.
Weather forecasting API
WeatherNext models on Google Cloud →
Google DeepMind's AI weather forecasts, available as a developer API and as raw forecast data in Earth Engine, BigQuery and Vertex AI, with ensemble scenarios out to 15 days.
Weather forecasting models
WeatherNext 2 →
Google DeepMind's ensemble weather and cyclone forecasting models, released with code, pretrained weights and runnable notebooks. Includes the checkpoint used operationally by the National Hurricane Center in 2025 and a one-degree Mini variant sized for a single GPU.
Web data API
Firecrawl →
A hosted API that crawls, scrapes and structures web pages into clean text for agents and retrieval pipelines, handling the JavaScript rendering and rate limiting you would otherwise build yourself.
Web publishing control
Cloudflare AI Bot Controls →
A free Cloudflare setting, live since July 1 2026, that lets any site separately allow or block three kinds of AI crawler: search indexers, live AI assistants, and model-training scrapers.
Web security
Anubis WebAssembly challenges →
An open-source anti-scraping proof-of-work system adding a faster WebAssembly, memory-hard argon2id challenge path with a no-WASM fallback.
Website bot policy
Cloudflare AI crawler controls →
Free-tier controls that split AI crawler traffic into Search, Agent and Training, each set independently to allow, block site-wide, or block only on ad-bearing pages. These are edge blocks on classified traffic, not robots.txt requests. New domains change default on September 15, 2026.
World model
Atlas (early access) →
World Labs' new omni world model for spatial intelligence: generates up to a minute of 1440p video along a camera path you specify exactly, reconstructs real scenes from two or three photos, and outputs explicit 3D. No public weights or API yet -- early access is by request.
Writing-style leaderboard
The Slop Index →
An open leaderboard scoring how much like generic AI prose a model writes, combining blind pairwise crowd votes with mechanical style measures against a pre-2022 human reference corpus. Methodology and generations are public; treat the rankings as a prototype, since the project's own published counts do not reconcile.