The Day in AI

• Google released Gemini Decision 1, which scores up to 1,000 closed-set questions per request for $0.10 per million input tokens.
• Alibaba opened a preview of Qwen4-Max, the first model from its Qwen4 family, with image and video input and a million-token context.
• Google DeepMind released AlphaGenome 2 and says it found 30% of causal variants on an immune-cell test, against 16% for the best alternative.
• Anthropic launched Claude Meetings, Whiteboards and Sheets in beta on 10 November and made Claude Data generally available.
• Mercor reportedly raised $750 million at a $20 billion valuation.

Gemini Decision 1 comes with the new Gemini Decisions API and handles yes or no, multiple choice and rubric-scoring questions. It scores the candidate options in a single forward pass and returns probabilities and confidence scores. Because it writes no answer text, Google charges zero output tokens. One request pairs a multimodal input of up to 256K tokens (text, images, audio, video or PDFs) with up to 1,000 independent questions.

Google reports end-to-end latency of 150 to 250 milliseconds, of which 50 to 100 milliseconds is model compute. Google hasn't said how the answers compare in accuracy with a generative Gemini model asked the same questions.

Alibaba opened a preview of Qwen4-Max, the flagship of a Qwen4 family that the Qwen team says is coming soon. It takes image and video input and up to a million tokens of context. The team says it improves knowledge, perception, coding and agentic work, but there are no benchmark figures yet. Developers reach it through the Token Plan subscription, which AIBase reports is discounted to as little as a tenth of the original price for a limited time, and it is in early use in the Qwen app, Qoder, QwenWork and QwenStudio.

Google DeepMind released AlphaGenome 2, a DNA model that predicts how genetic variants change gene activity in a given cell type or state. The first AlphaGenome produced fixed outputs for bulk tissues and cell lines. Version 2 takes the tissue, cell type or cell state as an input next to the DNA sequence and predicts one condition at a time, and a cell encoder lets scientists define a condition from their own single-cell data without retraining. DeepMind reports that on TenK10K, which covers 28 immune cell types, it identified 30% of causal variants against 16% for the best comparison model, and got the direction of the gene effect right 43% of the time against 23%. It was trained on over 107 million human and mouse cell profiles and is offered through a closed API. Some access details are only partly confirmed.

• Anthropic launched Claude Meetings, which joins Google Meet calls as a visible participant, along with Claude Sheets and Whiteboards in beta, and made Claude Data generally available on 10 November.
• Google DeepMind released Gemini Robotics 2.5 on 10 November and reports success rates 30% to 100% higher than earlier baselines across over 200 tasks.
• Baidu put ERNIE 5.5 Flash Preview, a 280B mixture-of-experts model with 14B active parameters, on its Qianfan API and promised open weights by 25 November.
• Apple is building Harbor, a Mac app that runs open-weight models locally and pools several Macs over Thunderbolt 5, according to a leak reported by 9to5Mac.
• MiniMax released weights for H3-Image, which it says makes 16-megapixel 4K images in 10 to 20 seconds with up to 16 reference images.
• Alibaba released Qwen Code 1.0 Preview, which moves the agent harness from Qoder into the open-source Qwen Code project.
• Santosh Janardhan, Meta's head of infrastructure, who joined the company in 2009, will leave in early 2027, and former Amazon executive Dave Brown takes over infrastructure.
• New York has an AI safety law that has taken effect and puts new oversight on OpenAI and Anthropic, TechCrunch reports.
• Fractile, a British AI inference chip startup, reportedly raised $720 million at a $7 billion valuation.
• OpenAI reportedly passed $10 billion in quarterly revenue, according to TechCrunch.