OpenAI also claims AI proofs of the Kakeya conjecture and Yang-Mills existence, and has handed all three results to outside expert groups for review.
The mid-sized model costs $1 per million input tokens, adds computer use and a 1.05M-token context, and replaces GPT-5.6 Terra.
Traditional models pointed Isaias toward Louisiana; the DeepMind model the National Hurricane Center now uses picked the western Florida Panhandle.
Magnus Lynch's 200-lesion test found the Inception v3 network named the right diagnosis first 62% of the time, versus 42% for both frontier LLMs.
The Apache 2.0 model from BOSS Zhipin's AI lab carries a 524K-token context window and serves as the teacher for a planned 3B distilled model.
Teams run by the 120B-parameter Avalon-M1-Mini won 14 of 16 benchmarks, but trained agents grew less likely to report wrongdoing to humans.
In a no-search follow-up, GPT still favored Democrats while Grok declined to pick in 27 of 30 races, pointing researchers to different roots for each lean.
Block says Dessa runs faster than real time on a MacBook Pro, was trained on a single Nvidia GB200 NVL72, and is opening first to AI safety researchers.
Artificial Analysis ranks the 2.8T-parameter model 4th on its Intelligence Index, and Moonshot says full weights will ship by November 4.
Built by Canva Research and trained on millions of templates, Craft renders 2K images with text in 55 languages and is rolling out to all Canva users.
The 14B Apache 2.0 model reads raw DNA in a chat prompt and scores within a point of GPT-6.1 Sol on DNA benchmarks.
Human-directed agents explored more than 100 candidate approaches; unguided AI runs with a similar token budget fell short, Microsoft says.
The 40B mixture-of-experts model activates 2.5B parameters per token and ships under Apache 2.0, with weights on Hugging Face.
Thousands of agents, each assigned one gene disruption, cited 700+ papers and produced leads human experts preferred 75% of the time over a single agent.
Anthropic's research model screened 553,052 design-toxin pairs, and two lead binders hit single-digit nanomolar affinity against the Mexican scorpion toxin Cn2.
The new SKU runs 100B-parameter models on device; two units cluster to 128GB and 200B via Nvidia Sync Cluster Assistant.
Two cryptanalysts used frontier LLMs to break wartime codes that had resisted human researchers for two decades.
Nearly 950 Claude agents worked 21 hours across 210 million tokens to flag a previously uncharacterized enzyme found in bacteriophages.
The new flagship chip's sensing hub supports models up to 200M parameters and a 30B MoE, outpacing Apple's 20B on-device model.
Higher cache hit rates, new diagnostics, and developer-set breakpoints aim to cut latency and API bills for GPT-6 workloads.
Output tokens drop to $20 per million from $25, and the new model outperforms the larger Fable on several coding and knowledge tasks.
The smaller siblings of GPT-6 Astra land with sharper factuality and a price cut, hours after Anthropic pushed Opus 5.5.
The GPU-accelerated ROS package hits 1.3M developers with agentic workflows, 5.5x faster pose tracking, and Ubuntu 24.04 support.
The model fills gaps in papyrus fragments and will be free for academics through a chatbot interface.
Nvidia's Halos system spans chips, OS, simulation and certification as physical AI moves from pilots into factories, roads and warehouses.
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
