Ground Truth.
AI, checked against the source.
← 2026-07-072026-07-08later →

Grok 4.5 arrives claiming Opus-class quality at a third the price

2026-07-08

SpaceXAI released Grok 4.5, a 1.5-trillion-parameter model priced at $2 per million input tokens and $6 per million output, undercutting frontier rivals roughly threefold while claiming comparable coding quality.

grok · xai · frontier-models · coding · pricing · model-launch

Chinese open models now handle a third of US enterprise AI traffic

2026-07-08

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.

china · open-weights · deepseek · glm · enterprise · economics · market-shift

Mistral's first robot model navigates unseen buildings with a single camera

2026-07-08

Mistral released Robostral Navigate, an 8-billion-parameter model that steers robots through buildings they have never seen using one ordinary RGB camera and a plain-language instruction, beating multi-sensor systems that rely on LiDAR and depth cameras.

mistral · robotics · navigation · embodied-ai · sim-to-real · model-launch

New world models generate depth and motion, not just video, to train robots

2026-07-08

A cluster of world-model papers led by RynnWorld-4D shows AI generating synchronized color, depth, and motion instead of flat video, so robots can practice manipulation inside a generated world and turn predictions directly into actions.

world-models · robotics · video-generation · depth · teleoperation · research

A new attention method learns what to ignore, and reads 64x more text

2026-07-08

Tencent's Hunyuan team introduced HiLS, a sparse-attention method that learns end-to-end which parts of a long document to focus on, matching full attention while handling context 64 times longer than it was trained on.

attention · sparse-attention · long-context · efficiency · tencent · research

OpenAI says a leading coding benchmark can no longer tell the best models apart

2026-07-08

OpenAI published an analysis concluding that SWE-Bench Pro, a widely-cited coding benchmark, has hit a roughly 70% noise ceiling where higher scores may reflect quirks rather than real skill, and retracted its recommendation to use the benchmark to rank frontier models.

openai · benchmarks · coding · evaluation · swe-bench · methodology

Google's Gemma 4 is a small open multimodal family that skips the image encoder

2026-07-08

Google released Gemma 4, an open-weight model family from 2.3 to 31 billion parameters that natively handles vision and audio, including a 12-billion-parameter variant that ingests raw image and audio patches with no separate encoder.

google · gemma · open-weights · multimodal · encoder-free · model-launch

A Brown professor moved a final in person, and the near-perfect scores collapsed

2026-07-08

A Brown University economics professor who suspected AI cheating switched his take-home final to an in-person exam, and average scores fell by roughly half — a stark data point in the growing fight over what take-home 'mastery' now measures.

education · cheating · academic-integrity · culture · chatgpt · society

Microsoft's Flint gives AI agents a reliable way to make charts

2026-07-08

Microsoft open-sourced Flint, a compact chart specification language that lets AI agents describe a visualization in structured JSON and compile it reliably to Vega-Lite, ECharts, or Chart.js, instead of hand-writing brittle plotting code.

microsoft · ai-agents · visualization · open-source · developer-tools · mcp

Anthropic switches Fable 5 to usage billing and turns on government ID checks

2026-07-08

Starting today, Anthropic bills its flagship Fable 5 model by usage at $10 per million input tokens and $50 per million output across all tiers, and its government-ID verification requirement for Fable 5 access takes effect as part of an export-control redeployment.

anthropic · fable-5 · pricing · governance · export-controls · policy

A repo collecting every major AI's hidden system prompt is topping GitHub

2026-07-08

A GitHub repository gathering the extracted system prompts of nearly every major AI product — from Claude and GPT to Gemini, Grok, and Cursor — surged up the trending charts, part of a week dominated by AI-agent infrastructure tooling.

github · system-prompts · ai-agents · open-source · transparency · developer-tools

← 2026-07-072026-07-08later →