Gemini Decision 1
A Google decision model that returns calibrated probabilities for yes/no, choice and rubric questions instead of text
Context window
262,144 tokens
Input / 1M tokens
$0.10
Output / 1M tokens
N/A (no generated text)
Provider
Priced per input token only: $0.10 per 1M for text, image, video, audio and PDF input, with 0 output tokens and no thinking tokens. The shared input is billed once per request however many questions it carries; each question adds only its own instruction and criteria tokens. Free tier in Google AI Studio and the Gemini API. No Batch API and no context caching. · Data verified 2026-11-17
Gemini Decision 1 (November 11, 2026) is a Google model that answers closed-set questions instead of generating text. Served through the new Gemini Decisions API, it scores candidate options in a single forward pass and returns calibrated probabilities plus a 0 to 1 confidence for yes/no, multiple-choice (2 to 255 options) and ordered-rubric questions. One request can carry up to 1,000 questions over a shared text, image, audio, video or PDF input of up to 262,144 tokens. Google reports 150 to 250 ms end-to-end latency that stays nearly flat from 1 to 100 questions, the lowest calibration error in its comparison (0.0250 Top-1 ECE vs 0.0494 for TypeSafe Jev 1.14 and 0.0721 for GPT-6 Luna) and nearly 5x lower cost than Gemini 3.5 Flash-Lite on the same tasks. Google says it already runs model routing in Gemini Notebook and a safety layer for MCP tools in Google Search.
How to access it
Gemini API Decisions endpoint (model ID gemini-decision-1) through Google AI Studio, with a free tier for prototyping; the SDK samples set api_version to v1beta. Google says support in Gemini on Google Cloud is coming soon and that integrations with OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode are rolling out.
Strengths
- ✓Returns calibrated probabilities and a confidence score, with zero output tokens
- ✓Lowest Top-1 calibration error in Google's comparison: 0.0250 vs 0.0494 (Jev 1.14) and 0.0721 (GPT-6 Luna)
- ✓Up to 1,000 questions per request, with the shared input billed once
- ✓Multimodal input: text, images, audio, video and PDF
- ✓150 to 250 ms end-to-end latency (Google-reported)
Best for developers who...
When to choose it (and when not to)
Reach for Gemini Decision 1 when...
- →You need many fast yes/no, choice or score decisions per request with calibrated confidence
- →Routing or guardrail steps where a generative call adds latency and output-token cost
- →Multimodal classification over images, audio, video or PDFs
Look elsewhere if...
- ✕Any task that needs generated text, explanations or tool calls
- ✕Open-ended questions without a fixed answer set
- ✕Batch jobs or cached prompts (neither is supported)
- ✕Lowest total price: in Google's own test TypeSafe Jev 1.14 cost $0.68 vs $0.88 on the 7 shared evals
How to use it
- ›Put shared context in input once and ask up to 1,000 questions against it
- ›Add true_description and false_description criteria to yes/no questions to pin down the decision boundary
- ›Use score questions with 2 to 10 rubric levels for grading
- ›Lower image and video resolution to cut token cost when fine detail is not needed
Quickstart
Pythonfrom google import genai
client = genai.Client(http_options={"api_version": "v1beta"})
result = client.decisions.create(
model="gemini-decision-1",
input="The blue mug I ordered on Monday arrived cracked on Wednesday.",
questions={
"is_damaged": {
"type": "noul",
"instructions": "Did the item arrive damaged?",
},
},
)
answer = result.answers["is_damaged"]
print(f"P(yes)={answer.noul:.2f} confidence={answer.confidence:.2f}")Install with pip install -U google-genai. Sample from Google's Decisions API docs; noul is the API's yes/no question type, and choice and score are the other two.
API model id: gemini-decision-1
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Top-1 ECE (calibration) | 0.0250 | Lower is better, 7 shared evals; TypeSafe Jev 1.14 0.0494, GPT-6 Luna (OpenAI Decisions API) 0.0721, Gemini 3.5 Flash-Lite 0.2104 |
| JevBench | 92.2% | Public split; Jev 1.14 90.0%, GPT-6 Luna 90.9%, Gemini 3.5 Flash-Lite 86.6% |
| Amazon ESCI | 58.7% | Macro-F1; Jev 1.14 56.7%, GPT-6 Luna 54.7%, Gemini 3.5 Flash-Lite 47.2% |
| Mind2Web Multimodal | 86.2% | Jev 1.14 86.8%, GPT-6 Luna 81.8%, Gemini 3.5 Flash-Lite 78.0% |
| When2Call | 79.9% | Jev 1.14 82.0%, GPT-6 Luna 70.4%, Gemini 3.5 Flash-Lite 60.9% |
| Global-MMLU-Lite | 86.0% | Non-English; Jev 1.14 85.0%, GPT-6 Luna 79.6%, Gemini 3.5 Flash-Lite 78.8% |
| Video-MME | 69.1% | No subtitles; Gemini 3.5 Flash-Lite 63.3% |
| Total price, 7 shared evals | $0.88 | Jev 1.14 $0.68, GPT-6 Luna $0.93, Gemini 3.5 Flash-Lite $4.14 (Interactions API, minimal thinking) |
Compare Gemini Decision 1
Compare Gemini Decision 1 with any other model
Build a comparison →All model comparisons →