The startup hit a $100M revenue run rate 10 months after first deployment and is now installing Blackwell systems at Foxconn with Nvidia.
The University of Pennsylvania team mines living and extinct genomes with GPT-powered tools to surface antimicrobial candidates against drug-resistant infections.
The AI music startup retires older models and pivots to licensed data from Warner, BMG and Believe as Sony and Universal lawsuits continue.
An internal model above GPT-6 Astra solved a 90-year-old problem, OpenAI says — a day after two researchers using its Codex published a related proof.
President Greg Brockman says the model may mark when AGI arrived, weeks after an earlier OpenAI system hacked Hugging Face.
The new model trains on live satellite data instead of physics simulations, cutting the six-hour lag that plagues traditional forecasting.
The Russian startup fused a 753B-parameter GLM-5.2 with a 4B Qwen-3.5 at one-twentieth the cost of the larger model.
The twinned 5.1 release cuts token costs, reduces false-positive refusals, and finally brings Zero Data Retention to Fable.
The AI upscaler carries a 50–60% performance hit and needs 6x frame generation to hit 1080p on midrange cards.
Code changes reviewed this week show Codex agents that 'continue working until put to sleep' and create their own follow-up tasks.
The Omni update pushes generative video to 4K, extends clips to 40 seconds, and cuts draft costs to a third at 360p.
Vera pairs with NVLink Fusion and a new NVHBM custom high-bandwidth memory tier aimed at trillion-parameter agent workloads.
The new audio model handles specialized jargon, tags up to three speakers, and lands while Gemini 3.5 Pro remains overdue.
The new speech-to-text model hits a 4.0% streaming WER, cuts time to final transcription by 70%, and supports over 85 languages.
Perceptron's open-weight Isaac 0.5 model targets warehouses and factories, trained on a million hours of video plus petabyte-scale robotic trajectories.
The reasoning-focused release adds agentic RL training on the larger variants and a 128,000-token context window across the family.
Claude will now carry context across chat and Cowork, and users can read, edit, or delete stored memories on any topic.
Apple's 2nm M6 and dual-die M5 Ultra push up to 512GB of unified memory and 1.2TB/s bandwidth — aimed squarely at developers running local LLMs.
The Broadcom-partnered chip ships in small volumes at the end of 2026, with broader deployment in 2027.
The new accelerator hits 3,400 output tokens per second on 100,000-token contexts, 4x the nearest platform, with Nebius as first cloud adopter.
A free reasoning model appeared on OpenRouter this week with no named creator, drawing praise from Stripe's Patrick Collison and speculation across the AI community.
The London lab, fresh off a $50M seed, says its DeepMind-alumni-built agent matches frontier systems using a fraction of the parameters.
New Nvidia research shows a custom harness with a supervisor agent lifted Claude Opus 5 from 30% to a perfect score on the interactive reasoning benchmark.
Google DeepMind extends 15 years of games research from Atari to EVE Online with a new agent that collaborates with human players.
The Cambridge startup founded by ex-DeepMind and Boston Dynamics researchers is chasing a GPT-3 moment for physical intelligence.
The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.
The briefing read inside teams at
