Top story
Biohub expands its AI-biology data initiative to $1.8 billion in commitments
The package combines new funding from Google DeepMind, Isomorphic Labs and Meta, a five-year Department of Energy commitment, and NIH datasets funded earlier. Reuters reports that commercial funders get early access to their datasets before public release; government-funded work has no such restriction.
Why it mattersCoordinating cell-response measurements across institutions gives researchers a path to the scale of data needed for predictive biology models.
Also
OpenAI begins rolling out GPT-6 with interactive answers in ChatGPT
Intelligent UI can put charts, forms and task-specific tools in conversations. GPT-6 Sol is rolling out to Plus, Pro, Business and Enterprise users; Free and Go users are scheduled to follow Thursday with GPT-6 Luna.
Why it mattersUsers can change inputs and explore results without leaving the conversation.
Anthropic releases Claude Haiku 5.5 at a lower API price
The new small model is available across its platforms at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens; Anthropic also halved Sonnet 5.5 cache-read pricing.
Why it mattersDevelopers running many short requests have a cheaper Claude tier.
Broadcom seeks more than $50 billion to finance OpenAI's custom chips
The Wall Street Journal reports that Broadcom has held early talks with Apollo and Blackstone about financing chips it is developing with OpenAI; the proposed size could change.
Why it mattersIf arranged, the funding would help OpenAI buy custom-chip capacity instead of relying solely on rented cloud hardware.
Research worth a look
Epoch AI's InnovationEval finds today's agents short of independent AI-algorithm research
In a small initial test, Epoch found GPT-5.6 Sol and Claude Fable 5 did not match a human-designed post-training method after correction for run selection and out-of-scope changes.
Tools & releases
UK AISI releases Transect
The open-source package turns long agent-evaluation transcripts into linked activity timelines that reviewers can check against the original exchanges.
Liquid AI opens two edge decision models
Its d1-3B accepts text and images, while the experimental d1-omni-600M accepts text with images or audio; the company's performance comparisons are vendor-run.
Google opens SynthID Detector worldwide in English
The website lets anyone check images, video and audio for SynthID watermarks from Google and participating partners, including OpenAI, Nvidia and Kakao; Apple support is planned. Its checks cover participating providers' watermarks, leaving other AI-made media outside its detection range.