Google launched the Gemini Decisions API on November 11, 2026, powered by a new model called Gemini Decision 1 that never writes text at all — it scores up to 1,000 yes/no, multiple-choice, or rubric questions about a single input in one pass and returns probabilities in about a quarter of a second. Why it matters: a whole category of “decision models” has formed in a few weeks, and Google just became its biggest player.
November 11, 2026
What happened
Google DeepMind introduced the Gemini Decisions API through the Gemini API in Google AI Studio. Instead of generating answer text token by token, Gemini Decision 1 (model name gemini-decision-1) scores your candidate options in a single forward pass. One request can carry a shared multimodal input — text, images, audio, video, and PDFs, up to 256K tokens — plus up to 1,000 independent questions, and every answer comes back with a confidence score from 0 to 1 so your code can route borderline cases to a fallback or a human reviewer.
The details
The API supports three question types: yes/no checks that return the probability a condition holds, multiple choice with 2 to 255 options returning normalized probabilities for every candidate, and ordered rubric scores that return a weighted grade. Google prices it at $0.10 per 1 million input tokens with zero output tokens billed, and says the shared input is billed once per request no matter how many questions you attach. Reported latency is 50–100 milliseconds of median model compute and 150–250 milliseconds end to end for contexts up to 16K tokens, staying nearly flat from 1 to 100 questions per request. Google says that makes it nearly 5x cheaper across its shared benchmark suite than Gemini 3.5 Flash-Lite used through the Interactions API. A no-cost tier is available in Google AI Studio for prototyping, and integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit, and OpenCode, with Gemini in Google Cloud support coming later.
How the benchmarks look — and the caveat
In Google’s own launch benchmarks, Decision 1 led five of the seven tests that all four rival models ran, scoring 92.2% on JevBench against GPT-6 Luna at 90.9% and Jev 1.14 at 90.0%, and it posted the lowest calibration error in the comparison at 0.0250 — roughly half Jev’s 0.0494. Those figures are Google-reported. The independent picture is more mixed: Artificial Analysis’s Decision Index v2.1 places Gemini Decision 1 ninth of 18 models at launch with a score of 65.98, behind Drex Max v1.1 (72.62) and Jev 1.14 (66.23), and Jev is cheaper for plain text work at $0.042 per million input tokens. Google itself already uses the API internally — as a real-time router in Gemini Notebook (which it says cuts latency up to 40% on everyday queries), in its Stitch design tool, and as a safety layer for MCP tools in Search.
Why it matters
Most of what AI agents actually do is not writing paragraphs — it is picking: which tool to call, which queue gets this ticket, does this photo match this claim, is this action safe. Doing that with a full generative model means paying for text you then have to parse back into a decision. Decision models skip that step, and the market now has entries from TypeSafe (Jev), OpenAI (its Decisions API, generally available October 22), Amazon (the open-weight Strands Decider), Cloudflare, Microsoft — and now Google, with the only natively multimodal option that handles audio and video in the same request. For builders, the practical takeaway is to test on your own checklist before believing any launch chart: the independent index and Google’s own numbers tell noticeably different stories.
Frequently asked questions
What is Gemini Decision 1?
It is a specialized Google model behind the new Gemini Decisions API. It does not generate text; it scores up to 1,000 yes/no, multiple-choice, or rubric questions about one shared input in a single pass and returns probabilities and confidence scores.
How much does the Gemini Decisions API cost?
$0.10 per 1 million input tokens, with no charge for output tokens. The shared input is billed once per request. A free tier for prototyping is available in Google AI Studio.
How is this different from asking a normal Gemini model?
A generative model writes an answer you then have to parse; a decision model returns structured probabilities directly, in 150–250 milliseconds end to end on inputs up to 16K tokens, according to Google. The trade-off is that it cannot explain itself or extract arbitrary text — those jobs still need a generative model.
Sources: Google (Gemini Decisions API launch post, Nov 11, 2026), VentureBeat, MarkTechPost, Unite.AI