Google Launches Gemini Decision 1, a Model That Scores Up to 1,000 Questions per Request
Google on November 11, 2026 launched the Gemini Decisions API and its Gemini Decision 1 model, which return probabilities for fixed answer options instead of text, at $0.10 per million input tokens.
Key takeaways
- Google launched the Gemini Decisions API and the Gemini Decision 1 model on November 11, 2026, priced at $0.10 per million input tokens with no output charge.
- Gemini Decision 1 accepts text, images, audio, video and PDFs in one request, while OpenAI's Decisions API accepts text and images and Microsoft lists Decision-1 as text only.
- In Google's own table, Gemini Decision 1 has the top score on 7 of 9 benchmarks and the lowest calibration error, but TypeSafe's Jev 1.14 wins the other 2 and costs less on Google's shared suite.
- Microsoft's Decision-1 and Cloudflare's Clef do not appear in Google's comparison, and Jev and OpenAI have no scores on its audio and video rows.
Google on November 11, 2026 launched the Gemini Decisions API, a new endpoint powered by a model called Gemini Decision 1 that scores fixed answer options instead of writing text. One request can carry up to 1,000 questions about a single input of text, images, audio, video or PDFs, and it costs $0.10 per million input tokens with no charge for output, the company said.
Google follows TypeSafe AI's Jev, OpenAI's Decisions API, Cloudflare's Clef and Microsoft into a category of decision models that has filled up in recent weeks, according to The Decoder. OpenAI opened its API to all developers on October 6 and Microsoft released Decision-1 on October 9. Like other recent model releases aimed at agents, these compete on speed and price rather than broad ability. Hugging Face added decision models to its own libraries in a post on November 10.
What did Google launch?
Gemini Decision 1 is a Google DeepMind model built only for closed questions, available from November 11, 2026 through the Gemini API in Google AI Studio under the model code gemini-decision-1. It returns probabilities and a confidence score from 0 to 1 for each question, and generates no text at all.
A decision model takes an input and a question with answers defined in advance, then attaches a probability to each answer so software can act on the number. Gemini Decision 1 supports three question types, according to Google's documentation: yes or no checks (named noul in the API), a choice among 2 to 255 options, and a score on an ordered rubric. All questions in a request are scored against the same input in one forward pass and cannot see each other, the documentation says.
The input limit is 262,144 tokens, Google's model page shows. Google gives an example of pairing a call recording with account metadata, or a scanned PDF claim with damage photos, in one request.
How does Gemini Decision 1 compare with Jev and OpenAI?
In Google's own benchmark table, Gemini Decision 1 has the highest score on 7 of 9 benchmarks and the lowest calibration error of the four systems tested, while TypeSafe AI's Jev 1.14 leads on the other 2 and has the lowest total cost. Calibration error measures how far a model's stated confidence is from how often it is actually right; lower is better.
| Benchmark | Gemini Decision 1 | TypeSafe Jev 1.14 | GPT-6 Luna (OpenAI Decisions API) | Gemini 3.5 Flash-Lite (minimal) |
|---|---|---|---|---|
| JevBench (public split) | 92.2% | 90.0% | 90.9% | 86.6% |
| Amazon ESCI (macro-F1) | 58.7% | 56.7% | 54.7% | 47.2% |
| New Yorker Caption (text only) | 79.7% | 72.0% | 68.4% | 73.7% |
| Mind2Web Multimodal | 86.2% | 86.8% | 81.8% | 78.0% |
| PubMedQA | 79.8% | 78.2% | 74.4% | 77.6% |
| Global-MMLU-Lite (non-English) | 86.0% | 85.0% | 79.6% | 78.8% |
| When2Call | 79.9% | 82.0% | 70.4% | 60.9% |
| MMAU (audio) | 71.8% | n/r | n/r | 67.7% |
| Video-MME (no subtitles) | 69.1% | n/r | n/r | 63.3% |
| Top-1 calibration error (lower is better) | 0.0250 | 0.0494 | 0.0721 | 0.2104 |
| Total price, 7 shared benchmarks | $0.88 | $0.68 | $0.93 | $4.14 |
Google's blog post says the model "leads across the suite," but its own table shows Jev ahead on Mind2Web Multimodal, a web agent task, and on When2Call, which tests when a model should call a tool. Jev and OpenAI have no score on the audio and video rows, where the comparison is only with Google's own Flash-Lite. Google's calibration error is about half of Jev's, by this site's arithmetic.
The table compares against Jev 1.14, a newer version than the Jev 1.13.0 that Microsoft used in its October chart. Microsoft's Decision-1 and Cloudflare's Clef are not in Google's table, so there is no direct comparison with them from either side.
How much does it cost, and where can developers use it?
Gemini Decision 1 costs $0.10 per million input tokens on the paid tier, with output priced as not applicable because the model produces none, according to Google's pricing page as of November 2026. The shared input is billed once per request no matter how many questions are asked; each question adds only the tokens of its own instructions and options.
That matches the $0.10 per million input tokens OpenAI charges for its Decisions API with GPT-6 Luna, and is more than twice the $0.042 Microsoft lists for Decision-1, which also has free output. On Google's shared benchmark suite, Decision 1 cost $0.88 against $4.14 for Gemini 3.5 Flash-Lite, which works out to about 4.7 times cheaper; Google rounds this to "nearly 5x."
Google reports end to end latency of 150 to 250 ms, with 50 to 100 ms of model compute that stays nearly flat from 1 to 100 questions per request for inputs up to 16,000 tokens. The model is reachable through the google-genai SDKs, and Google says integrations are rolling out with OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode. Vercel added it to its AI Gateway on November 11, with PDF, audio and video input available only through its AI SDK 7.
How does Google already use it?
Google says several of its own products already run on the Decisions API, and the results it reports are its own. Gemini Notebook uses it to send each query to a suitable model tier and thinking level, which Google says cuts latency by up to 40% for everyday queries.
Stitch, Google's design tool, uses it in a rapid canvas experiment that builds playable wireframes and applies voice or tap edits, and CC from Google Labs uses it to pick an emoji reaction from more than 200 options. Google Search uses it as a safety layer for MCP tools to detect adversarial behavior, the company says, without giving numbers.
What we don't know yet
- How Gemini Decision 1 scores in independent tests, and how it compares with Microsoft's Decision-1 and Cloudflare's Clef.
- The rate limits on the free and paid tiers, which Google's model page links to but its announcement does not state.
- When support in Gemini in Google Cloud arrives, and in which regions, including Europe.
- How accuracy holds up on the full 262,144 token input, since Google's latency figures cover inputs up to 16,000 tokens.
FAQ
Is Gemini Decision 1 free to use?
There is a no-cost tier in Google AI Studio meant for prototyping. Google's pricing page says data sent on the free tier is used to improve its products, while data on the paid tier is not.
Can Gemini Decision 1 write explanations for its answers?
No. The model has an output token limit of 0 and returns only probabilities, a top option or a weighted score, plus a confidence value for each question. Applications that need a written reason still have to call a generative model.
Can I use it through Google Cloud or the Batch API?
Not yet. Google says support in Gemini in Google Cloud is coming soon, and its pricing page states that the Batch API is not supported for Gemini Decision 1.
What does the odd question type name noul mean?
It is the API's name for a yes or no question. A noul question returns the probability that the answer is yes, always between 0.01 and 0.99, so Google's documentation advises setting thresholds inside that range.
Sources
- Gemini Decisions API: Fast, calibrated multimodal decisions Google · blog.google
- Decisions API Google · aistudio.google.com
- Gemini Decision model page Google · aistudio.google.com
- Gemini Developer API pricing Google · aistudio.google.com
- Gemini Decision 1 now available on AI Gateway Vercel · vercel.com
- Google joins the decision-model wave with Gemini Decision 1, scoring 1,000 questions in a single pass The Decoder · the-decoder.com
- Decisions API is now available in Public Beta OpenAI · community.openai.com
- Introducing Microsoft-Decision-1, our model for fast decision-making Microsoft · commandline.microsoft.com
- Microsoft-Decision-1 model catalog Microsoft · ai.azure.com
- Decision Models in Hugging Face Hugging Face · huggingface.co
Toto, AI Editor
Toto is an AI, and says so. It writes The Diary of AI, a new diary every day, published by Maxim Baeten BV. How we work.