Google's Gemini Decisions API Scores Options in One Pass for $0.10 per Million Input Tokens
A.I. / news
Google's Gemini Decisions API Scores Options in One Pass for $0.10 per Million Input Tokens
Gemini Decision 1 returns probabilities instead of text, with no output-token charge and a benchmark lead Google has not backed with published scores.

Google DeepMind released Gemini Decision 1 on Nov. 11, a model that answers closed-set questions in one forward pass and bills $0 for output tokens. It sits behind a new Decisions API in the Gemini API and Google AI Studio.
Google's launch post is by Min Choi, a product manager at Google DeepMind, and Felix Yu, a principal scientist there. The model ID is gemini-decision-1. Google's post does not say whether the API is in preview or generally available.
What the Decisions API returns
Instead of generating text, the model scores options Google supplies. One request can carry a single multimodal input of up to 256K tokens, spanning text, images, audio, video and PDFs, plus up to 1,000 independent questions. Each answer comes with a confidence score from 0 to 1.
| Question type | What you send | What comes back |
|---|---|---|
| Yes/No | A condition, optional true and false criteria | Probability the condition holds |
| Multiple choice | 2 to 255 options | The winner and normalized probabilities |
| Ordered rubric | Weighted levels starting at 0 | A weighted score and per-level probabilities |
Google lists the intended uses as tool selection, policy checks, model routing and context grading. The post says the model "scores your candidate options in a single forward pass" rather than writing tokens one at a time.
Price and latency, as Google states them
Google's price is $0.10 per 1M input tokens and nothing for output. The post says there are "no output-token charges, variable thinking-token bills, or separate audio transcription and OCR costs." A shared input is billed once per request, and each question adds only its own instruction and criteria tokens. Google AI Studio carries a no-cost tier for prototyping.
Latency is 150 to 250 ms end to end, with 50 to 100 ms of median model compute, according to Google. The post says latency stays nearly flat from 1 to 100 questions per request on contexts up to 16K tokens. These are vendor-supplied figures and the post names no outside party that measured them.
Benchmarks: a claim without scores
Google says Gemini Decision 1 "leads across the suite, with the lowest calibration error" on benchmarks covering agent routing, reasoning, ranking and multimodal understanding. It also says the model is nearly 5x cheaper than Gemini 3.5 Flash-Lite used through the Interactions API on the same suite.
The post's text gives no scores. The figures appear only in charts. Until an outside group runs the model, the leads-the-suite claim is Google's alone.

Who is using it, and where it is available
Google names four uses of its own. Gemini Notebook uses it for model routing, with up to 40% lower latency on everyday queries. Google Search uses it as a safety layer for MCP tools that detects adversarial behavior. Google Labs' CC picks an emoji reaction from more than 200 options, and Stitch uses it for canvas wireframing.
Outside Google, Vercel's changelog of Nov. 11 lists the model as google/gemini-decision-1 on its AI Gateway. Vercel says decision calls appear in logs and count toward budgets alongside other model requests. Google lists integrations rolling out with OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode.
The post shows sample code on api_version v1beta with the google-genai SDK. Support in Gemini in Google Cloud is listed as "coming soon," with no date. For other agent-side plumbing we have covered, see Pomerado's website-to-MCP pricing and Alibaba's OpenCodeReview token claims.
Sources
More in A.I.
- 01NVIDIA Nemotron 3.5 Bolt Is Called 4GB; Its 4-Bit File Is 6.18 GBThe Gemma 4 E4B derivative wins seven of eight benchmark rows against its base on NVIDIA's own harness, but loses five of eight to Qwen3.5 4B.
- 02Chrome Built a Fake Web to Test Auto Browse Against Prompt InjectionTwo Chrome AI Security engineers describe a gVisor sandbox that serves generated sites at real URLs, with injections chosen by a request header.
- 03Anthropic Prices IPO at $83, and Two Outlets Disagree on the ProceedsFortune reports about $79.4 billion raised on 980 million shares; The Next Web, citing Bloomberg and the FT, reports at least $68 billion on 801.4 million.
- 04Tencent Hy4-flash Card Says Weights Are Out; Launch Coverage Says PlannedThe Hugging Face page lists 220B parameters and a custom licence, while launch coverage gives 224B and a future release, at $0.070 per million input tokens.