deepseek
Multi-head latent attention: compressing the memory that inference actually runs out of Lesson
Multi-head latent attention compresses the key-value cache that language models must hold in GPU memory during generation, projecting keys and values into a small shared latent vector instead of storing them per head. DeepSeek introduced it in DeepSeek-V2, reporting a cache reduction of more than 90% with quality matching full attention.
DeepSeek gave its cheapest model eyes and did not change the price News
DeepSeek shipped an experimental vision version of its V4-Flash model that accepts images by base64, URL, or file upload, and bills it at exactly the same rate as the text-only model.
DeepSeek shipped an agent runtime where even the loop is a plugin News
DeepSeek published deepseek-harness, an MIT-licensed agent runtime in which models, tools, skills, sessions, sandboxes, storage, scheduling, the interface, and the agent loop itself are all plugins that can be swapped from configuration.
DeepSeek gave its cheap model eyes, then capped them at 384 tokens an image News
DeepSeek released deepseek-v4-flash-vision-exp, an experimental multimodal version of its cheapest model that accepts images directly in the same agent loop as text, but budgets each image to at most 384 tokens after resizing toward roughly 800 by 800 pixels.
DeepSeek is selling a checkpoint it has not published News
DeepSeek's API now serves a model version named DeepSeek-V4-Pro-0813 and at least five commercial hosts resell it by that exact name, but the company has not published a matching dated weights page, and none of the resellers undercuts DeepSeek's own price.
DeepSeek starts charging rush-hour prices on August 17 News
DeepSeek is replacing flat API pricing with peak and off-peak rates on August 17, and the steepest change hits cached input on its Pro model, which goes up twelvefold during Beijing business hours.
Chinese models passed American ones in OpenRouter traffic in June News
OpenRouter's own analysis dates the crossover where Chinese models overtook American ones in token share to early June 2026, driven by DeepSeek V4 Flash taking 70 percent of DeepSeek's agentic traffic.
DeepSeek's new open model is 1.6 trillion parameters and runs 49 billion of them per token News
DeepSeek published DeepSeek-V4-Pro on Hugging Face with 1.6 trillion total parameters, 49 billion activated per token, and a one-million-token context window, making it the largest openly downloadable model of the current frontier wave.
The harness, not the model, moved DeepSeek's score by twenty tasks News
Independent benchmarker Harrison 'sentdex' Kinsley re-ran DeepSeek V4 Flash 0731 on the same 89-task terminal benchmark under a different agent harness and watched it go from 44 solved to 64 solved, without changing the model.
DeepSeek warns of a significant API price rise, five days after being called 100 times cheaper News
DeepSeek added a footnote to its official pricing page warning that it plans to raise API prices significantly in the near future with no figure and no date attached, five days after an independent benchmark study priced its model at roughly 100 times less per task than Western frontier models.
The Cheap 284B Rig Is Really 768GB of Server Memory News
A builder running DeepSeek V4-Flash at 33 tokens a second on two RTX 3090s is holding about 6.6GB of weights per card and roughly 170GB per instance in system memory on a four-socket enterprise server, which is where the model actually lives.
llama.cpp shipped DSpark for DeepSeek V4 Flash, and almost everyone called it the wrong name News
llama.cpp release b10228 merged speculative-decoding support for DeepSeek V4 Flash, but the new 0731 checkpoint embeds DSpark and ships no MTP at all, so the widely repeated "MTP support landed" advice points users at a head their model does not contain.
Quantizing V4 Flash's KV cache in llama.cpp changes which tokens it picks News
A community experiment found that llama.cpp's generic 8-bit KV cache leaves DeepSeek V4 Flash's average perplexity almost unchanged while altering which tokens make the model's shortlist about one time in eight, because the model's attention makes discrete block-retrieval decisions that a small numerical error can flip.
DeepSeek's low effort setting writes more than its high setting, because the dial is just a prompt News
A reproduction across both a local copy and DeepSeek's hosted API found V4 Flash consuming substantially more tokens on its low reasoning-effort setting than on high, and the released encoder explains why: low injects no instruction at all while high prepends a paragraph demanding exhaustive deliberation.
A 284-billion-parameter model with a 3-gigabyte working set, and a 96-gigabyte disk bill News
An open-source engine called Mference runs DeepSeek V4 Flash on a 24 GB Mac with an effective memory footprint of about 3 gigabytes by streaming each token's experts off the SSD, but the checkpoint still occupies 90 to 98 gigabytes of disk and the test ran at a 4,000-token context.
llama.cpp ships the fix that lets DeepSeek V4 Flash call tools mid-thought News
DeepSeek's new open-weight model shipped without a standard chat template, and local agents kept dying when it emitted a tool call inside its reasoning; llama.cpp release b10217 and KoboldCpp v1.118 both landed fixes on 1 August.
DeepSeek re-trained V4 Flash without touching the architecture and its coding-agent score went from 7 to 54 News
DeepSeek published new MIT-licensed weights for V4 Flash on July 31 that change only the post-training, lifting the model's score on a real-world software-engineering agent test from 7.3 to 54.4 out of 100.
DeepSeek V4 Flash hits 32 tokens a second on one desktop News
A published benchmark shows DeepSeek's 284-billion-parameter V4 Flash generating 32 tokens per second entirely on one AMD Strix Halo machine, using aggressive quantization, speculative decoding and reduced expert routing.
Chinese open models passed US models in OpenRouter token share News
OpenRouter's own usage data shows Chinese models overtaking US models in token volume in early June, with DeepSeek roughly doubling its share to 18% - driven by token-hungry agent workloads routing to the cheapest capable endpoint.
DeepSeek paused its funding round after a leaked meeting transcript went viral News
Bloomberg reports DeepSeek told prospective backers it would not sign expected agreements for now, a suspension its sources attribute in part to viral posts about a leaked investor-meeting transcript attributed to founder Liang Wenfeng.
DeepSeek is designing its own AI chip -- and raising outside money for the first time News
Chinese AI startup DeepSeek is developing its own chip aimed at running trained models rather than training them, and is simultaneously raising its first-ever outside capital -- about $7 billion at a $52-59 billion valuation.
Chinese open models now handle a third of US enterprise AI traffic News
US companies now route more than 30% of their AI tokens through Chinese open-weight models like DeepSeek and GLM-5.2 every week since February, peaking near 46%, up from an 11% average the year before, according to CNBC's analysis of OpenRouter data.
DeepSeek's new open models give everyone a million-word memory by default News
DeepSeek previewed two free-to-download V4 models that can read a million tokens at once, no longer as a premium add-on but as the standard setting.
The trick that makes AI type faster just hit the top of Hacker News News
A small model guesses ahead and a big model checks the work in parallel - and this week two efforts pushing that idea, DeepSeek's DSpark and JetSpec, lit up the front page while the community argued over whether it's truly 'lossless.'
Frontier AI is getting more expensive while open models keep getting cheaper News
Closed frontier models are raising prices and tightening access just as Chinese open-weight models slash theirs, a structural reversal with big consequences for who builds with AI.
Suddenly, downloadable AI models look like an insurance policy News
With a top hosted model pulled overnight, a flood of powerful open models you can run yourself -- and run fast -- is being reframed from hobby to risk management.
llama.cpp b10228 Tool
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 b10217 Tool
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.
Unsloth DeepSeek-V4-Flash-0731 GGUF Tool
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.
SGLang Tool
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.
KoboldCpp v1.118 Tool
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.
DeepSeek-V4-Pro Tool
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.
DeepSeek-V4-Flash-0731 GGUF (Unsloth) Tool
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.
DeepSeek-V4 (Pro & Flash) Tool
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.
DeepSeek V4-Flash vision API Tool
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.
DeepSeek V4 Flash Vision (experimental) Tool
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.
DeepSeek V4 Flash 0731 (API) Tool
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.
DeepSeek V4 Flash 0731 Tool
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 Tool
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.
DeepSeek Harness Tool
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.