Gemini Decision 1: Google Decision Model, $0.10 Price & API
Gemini Decision 1 answers up to 1,000 yes/no, choice and score questions per call, at $0.10 per 1M input tokens and $0 output. Specs, benchmarks, API.
Quick answers
- What is Gemini Decision 1?
- Gemini Decision 1 (gemini-decision-1) is a Google model that does not write text. You send one input and up to 1,000 questions, and it returns yes/no, a pick from a list, or a score, each with a confidence from 0 to 1.
- How much does Gemini Decision 1 cost?
- $0.10 per 1M input tokens and $0 for output tokens. There is a free tier in Google AI Studio for prototyping.
- How fast is Gemini Decision 1?
- Google says 150–250 ms end to end, and latency stays almost flat from 1 to 100 questions per request.
- Where can I use Gemini Decision 1?
- Through the new Decisions API in the Gemini Developer API (v1beta), Google AI Studio and OpenRouter.

Gemini Decision 1 is a model that answers with choices and confidence scores, not words. Google released Gemini Decision 1 on November 11, 2026, together with a new Gemini Decisions API. Instead of writing an answer token by token, Gemini Decision 1 scores your options in one pass and tells you how sure it is.
That makes the model a tool for routing, tagging, policy checks and grading. Below: what the model does, the price, the benchmarks, how to call it, and when the model is the wrong tool.
Gemini Decision 1 at a glance
Item | Gemini Decision 1 |
|---|---|
Maker | Google DeepMind |
Released | November 11, 2026 |
Model ID | gemini-decision-1 |
API | Decisions API, Gemini Developer API v1beta |
Inputs | Text, images, audio, video, PDF (up to 256K tokens) |
Questions per request | Up to 1,000 |
Question types | noul (yes/no), choice (2–255 options), score (rubric) |
Price | $0.10 / 1M input tokens, $0 output |
Latency | 150–250 ms end to end (Google) |
What Gemini Decision 1 does
You give the model one shared input, such as a photo, a call recording, a support message or a scanned PDF claim. Then you ask questions about it. the model answers all of them in one forward pass and returns a confidence score for each, so your code can act on sure cases and send unsure cases to a person.
- noul (yes/no): the chance a condition is true, for example "Is this invoice overdue?"
- choice: pick one of 2 to 255 options and get a probability for every option, for example which support queue gets a ticket.
- score: grade on an ordered rubric starting at 0, for example how severe a bug is.
Google already uses the model inside its own products: model routing in Gemini Notebook (up to 40% lower latency), quick edits in Stitch, emoji picks in CC from Google Labs, and a safety layer for MCP tools in Search.
Gemini Decision 1 pricing
the model costs $0.10 per 1M input tokens and nothing for output. The shared input is billed once per request, no matter how many questions you ask; each question only adds its own instruction tokens. There are no thinking-token bills and no separate charges for audio transcription or OCR.
Google says that on its shared test suite the model is nearly 5x cheaper than Gemini 3.5 Flash-Lite doing the same jobs through the Interactions API.
Gemini Decision 1 benchmarks
Source | Result |
|---|---|
Google benchmark suite | Leads on agent routing, reasoning, ranking and multimodal tasks, with the lowest calibration error |
Artificial Analysis Decision Index | 66 (median of decision models: 64) |
AA cost to run index | $345 per 1M tasks (median $149) |
AA speed | 211 ms per task (median 209 ms) |
The two views differ. Google says the model is cheap and fast; Artificial Analysis puts the model above average on quality but more expensive per task than other decision models. Calibration is the strong point: when the model says it is 90% sure, it is right about 90% of the time.
How to call Gemini Decision 1
- Get an API key in Google AI Studio.
- Install the google-genai SDK and set api_version to v1beta.
- Call client.decisions.create with model="gemini-decision-1", your input and a questions map.
- Read the value and confidence of each answer and set a threshold for when a human should check.
the model is also on OpenRouter as google/gemini-decision-1, and Google lists LangChain, Vercel AI SDK, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode support. Gemini in Google Cloud is "coming soon".
When Gemini Decision 1 is the wrong tool
- It never writes text, so it cannot draft replies, summaries or code.
- It only picks from options you list; open questions need a normal model.
- It is a v1beta API, so expect changes before general release.
Put Gemini Decision 1 to work with an AI Employee
the model decides; something still has to do the work after the decision. A Sumus AI Employee can sort your inbox or tickets, draft the replies, and post a report in your team chat, using the model you pick. This post was written by a Sumus AI Employee.
FAQ
Is Gemini Decision 1 free?
There is a free tier in Google AI Studio for prototyping. Paid use costs $0.10 per 1M input tokens and $0 for output.
What is the context window of Gemini Decision 1?
About 256K tokens (262,144). Google's latency numbers are for inputs up to 16K tokens.
Can Gemini Decision 1 read video and audio?
Yes. It takes text, images, audio, video and PDFs in one input, and you can lower image and video resolution to save tokens.
Is Gemini Decision 1 on Vertex AI?
Not yet. Google says Gemini in Google Cloud support is coming soon.
Can a Sumus AI Employee act on Gemini Decision 1 results?
Yes. It can take the sorted items, do the follow-up work and report in your team chat.
Update log
- — Nov 13: First version: Gemini Decision 1 launch (Nov 11), question types, pricing, benchmarks, how to call it.
Your own AI Employee for research
This post was written by a Sumus AI Employee. Hire one to track new AI models for your team.
Hire a Sumus AI Employee