Google launches Gemini Decisions API for closed-set scoring
November 11: The Decoder adds that integrations are still rolling out, Google claims the lowest Top-1 ECE, and OpenAI, Cloudflare, and Microsoft recently shipped similar models.
Google’s Gemini Decisions API scores up to 1,000 closed-set questions in one pass, with probabilities returned in 150–250 ms.
Google DeepMind introduced the Gemini Decisions API, powered by Gemini Decision 1 (gemini-decision-1), in the Gemini API and Google AI Studio. One request can score yes/no, multiple-choice, and rubric or scale questions over text, images, audio, video, and PDFs, returning probabilities and confidence rather than generated text. Google says end-to-end latency is 150–250 ms and pricing is $0.10 per million input tokens with output tokens free, nearly five times cheaper than Gemini 3.5 Flash-Lite. Internal uses include Gemini Notebook routing, Stitch, and Search safety checks. The Decoder reports integrations are still rolling out and that Google claims the lowest Top-1 ECE calibration error.
- Input can mix modalities up to 256K tokens and 1,000 questions.
- Choice questions cover 2 to 255 options, each with a probability.
- Google cites 50–100 ms of model compute.
- Search uses it to flag adversarial behavior against MCP tools.
- A no-cost prototyping tier is available in Google AI Studio.
- OpenRouter, LangChain, and Vercel integrations are still rolling out.
What each outlet adds
beyond the shared facts- Google · AIToken limits, option counts, compute latency, and internal product uses.
- The DecoderRolling-out integrations, Top-1 ECE claim, and rival decision models.
- Vendors
- Organizations
- Google DeepMindOpenAICloudflareMicrosoft
Coverage
oldest first · one row per article · UTCGoogle launches Gemini Decisions API and Gemini Decision 1
One request can score up to 1,000 closed-set questions for $0.10 per million input tokens, with output tokens free.
Google launches Gemini Decision 1 for batch closed-ended scoring
The Decisions API scores up to 1,000 questions in one pass and returns probabilities in 150 to 250 milliseconds.