What happened?
On November 11, 2026, Google introduced Gemini Decisions API, a new developer interface designed for applications that need fast, structured decisions rather than lengthy AI-generated responses.
The API is powered by Gemini Decision 1, identified by the model name gemini-decision-1.
Unlike conventional generative AI models that produce text token by token, this system evaluates predefined options and returns probabilities and confidence scores.
Google is positioning it for tasks such as tool selection, model routing, policy checks, content classification and retrieval-context evaluation.
The API is available through the Gemini API and Google AI Studio.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
1. How the Decisions API works
Traditional AI applications often use a large language model to answer a question and then parse the response to extract a classification or decision.
For example, a SaaS application might ask an AI model whether a support ticket should be classified as billing, technical support or account management.
The model generates text, which the application then interprets.
Gemini Decisions API takes a different approach.
Developers define the available decision types and provide the relevant input. The model evaluates the options and returns structured results.
This is particularly useful when an application does not need an open-ended explanation.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
2. Three types of decisions
Google introduced three main decision primitives.
| Decision type | Purpose | Example |
|---|---|---|
| noul | Yes/no decision | Is this transaction suspicious? |
| choice | Multiple-choice selection | Which support department should receive this ticket? |
| score | Ordered rubric scoring | How severe is this reported issue? |
The multiple-choice format supports between 2 and 255 options.
The scoring format evaluates inputs against an ordered rubric, returning a weighted score and probability distribution across levels.
The results include confidence information that developers can use to determine whether a decision should be accepted or sent for additional review.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
3. Multimodal processing and up to 1,000 questions
One of the most notable capabilities is the ability to evaluate multiple questions against shared multimodal input.
Google says the API supports:
- Text.
- Images.
- Audio.
- Video.
- PDF documents.
A request can contain up to 256K tokens of input context and up to 1,000 independent questions.
For example, an insurance application could submit a scanned claim document, photographs of property damage and a customer call recording.
It could then ask multiple questions about missing information, claim categorization, potential inconsistencies and review priority.
The model evaluates the shared input in a single forward pass instead of generating separate conversational answers for every question.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
4. Pricing and performance
The pricing model is another important part of the announcement.
Gemini Decision 1 pricing
| Published specification | Value |
|---|---|
| Per 1 million input tokens | $0.10 |
| Output-token charges | $0 |
| Questions per request | Up to 1,000 |
| Maximum context | 256K tokens |
| Google-reported end-to-end latency | 150–250 ms |
Performance and pricing are Google's published specifications, not independently measured results.
Google says the shared input is billed once per request, while each question adds its own instructions and criteria tokens.
The company also claims nearly five times lower cost than Gemini 3.5 Flash-Lite with its Interactions API across a shared benchmark suite.
There is an important qualification: Google's detailed latency comparison describes contexts up to 16K tokens and near-flat latency for 1–100 questions. The 150–250 ms figure should not automatically be assumed for every 1,000-question, 256K-token request.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
5. How developers can use it
The API could be useful in several real-world applications.
SaaS customer support: Automatically categorize incoming tickets and determine whether human support is required.
AI agent orchestration: Select an appropriate model or tool based on the task rather than sending every request to an expensive reasoning model.
Content moderation: Evaluate predefined policy conditions and route uncertain cases for manual review.
Document processing: Extract structured decisions from invoices, applications, scanned forms and other documents.
Security workflows: Evaluate suspicious inputs and determine whether an action requires additional validation.
For companies building AI-powered products, this could provide a dedicated decision layer alongside their existing generative models.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
6. Ecosystem integrations
Google says integrations are rolling out across several developer platforms, including:
- OpenRouter.
- LangChain.
- Vercel AI SDK.
- LiteLLM.
- Pydantic AI.
- Pipecat.
- LiveKit.
- OpenCode.
Google Cloud integration is listed as coming soon rather than generally available through this announcement.
Google also describes internal use cases involving model routing in Gemini Notebook, rapid design iteration in Stitch and security checks in Search. These are company-reported applications, not independent performance evaluations.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
7. Limitations and what developers should consider
A confidence score should not automatically be interpreted as a guarantee of correctness.
Developers should evaluate the model against their own labeled datasets, establish confidence thresholds and define fallback behavior.
For sensitive workflows such as payments, account access, fraud prevention and compliance, low-confidence or high-impact decisions should be routed to human reviewers or additional validation.
The API also does not replace general-purpose reasoning models. It is designed for bounded decisions, not complex open-ended tasks requiring extensive explanations.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
What to watch next
The next important developments include independent evaluations of its calibration, SDK ecosystem adoption, Google Cloud availability and real-world production costs.
The broader development is the emergence of specialized AI decision models as a separate layer in agentic software architectures.Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
Why this is trending
Gemini Decisions API — Hacker News
Front page
| Community metric | Research snapshot |
|---|---|
| Points when checked | 242 |
| Comments when checked | 78 |
November 11 discussion snapshot.
Developers debated probability calibration, classification versus generative AI, token economics and practical use cases such as model routing, support-ticket filtering and human-review gates.
VentureBeat also independently covered the launch, focusing on the API's ability to score many questions against shared input. No verified same-window Reddit or X engagement was established.Community — Hacker News (11 November 2026): Gemini Decisions API; counts in this article are briefing snapshots.
Sources
- Google — Gemini Decisions API: Fast, calibrated multimodal decisions — November 11, 2026. Official launch, pricing and model capabilities.
- Gemini Decisions API Documentation — API documentation and developer implementation reference.
- VentureBeat — Google's Gemini Decision 1 model scores up to 1,000 questions in one API call — November 11, 2026. Independent coverage.
- Hacker News — Gemini Decisions API — November 11, 2026. Developer discussion.
Further reading
- Primary — Google (11 November 2026): Gemini Decisions API: Fast, calibrated multimodal decisions. Official launch, pricing and model capabilities.
- Primary documentation — Gemini Decisions API: API documentation and developer implementation reference.
- Independent — VentureBeat (11 November 2026): Google's Gemini Decision 1 model scores up to 1,000 questions in one API call.
- Community — Hacker News (11 November 2026): Gemini Decisions API; counts in this article are briefing snapshots.
Building a product in this space?
MORE FROM THE NEWSROOM
Related news
Artificial IntelligenceUS Moves to Regulate Remote Access to Advanced AI Chips Through Overseas Data Centers
The US Department of Commerce has released an interim final rule targeting remote access to advanced American semiconductors through overseas data centers. The measure introduces licensing requirements for certain high-volume computing arrangements serving companies headquartered in restricted destinations and adds destination-based controls for advanced chip exports to five countries.
Artificial IntelligenceOpenAI Launches ChatGPT Wallet, Letting AI Agents Complete Purchases With User Approval
OpenAI has introduced ChatGPT Wallet, a payment feature that allows its AI agents, called Dots, to complete online purchases after users review and approve each transaction. Developed with American Express, Mastercard and Visa, the initial rollout is limited to eligible US ChatGPT Pro subscribers.
Artificial IntelligenceCerebras Raises 2026 Revenue Outlook After 55% Quarterly Growth, but GAAP Losses Widen
AI hardware company Cerebras reported record third-quarter revenue of $210.8 million and raised its full-year 2026 revenue outlook. Its cloud and inference business grew rapidly, but the results also revealed a much larger GAAP net loss, highlighting the financial challenges of scaling advanced AI infrastructure.
