Gemini Decision 1 AI dashboard

Gemini Decision 1: 7 Key Features of Google’s New AI Decision Model

Google introduced Gemini Decision 1 on November 11, 2026, a new artificial intelligence model designed to make fast, structured decisions from text, images, audio, video, and PDF documents.

The model powers Google’s Gemini Decisions API, which is available through the Gemini API in Google AI Studio. Instead of generating lengthy responses, it evaluates predefined questions and returns selected answers, probabilities, and confidence scores.

Google says developers can submit up to 1,000 independent questions about shared input in a single request. The service costs $0.10 per million input tokens, with no output-token charges.

This development introduces a different approach to AI-powered automation. Rather than asking a general-purpose chatbot to explain everything, developers can use a specialized model to make specific decisions within their applications.

Key Takeaways

  • Google announced Gemini Decision 1 on November 11, 2026.
  • It powers the Gemini Decisions API.
  • The API supports text, images, audio, video, and PDFs.
  • Developers can evaluate up to 1,000 questions in one request.
  • It supports yes/no decisions, multiple-choice questions, and rubric scoring.
  • Google lists pricing at $0.10 per million input tokens, with no output-token charges.
  • The model is designed for structured decisions rather than general-purpose content generation.

What Is Google Gemini Decision 1?

Gemini Decision 1 is a Google AI model designed to evaluate information and return structured decisions.

It powers the Gemini Decisions API, which allows developers to submit information alongside specific questions. The model evaluates the available evidence and returns answers with associated probabilities and confidence scores.

Traditional generative AI models produce responses by generating text token by token. Gemini Decision 1 takes a different approach: it scores predefined options in a single forward pass rather than generating a conventional written response.

For example, an application could provide a scanned document and ask whether it contains a required signature. The model can evaluate that question and return a structured result.

Google introduced the model to support applications that need fast decisions, including model routing, policy checks, tool selection, and content evaluation.

1. It Can Evaluate Up to 1,000 Questions in One Request

One of Gemini Decision 1’s main features is its ability to evaluate multiple questions against the same input.

Google says developers can submit up to 1,000 independent questions in a single request.

This means an application does not necessarily need to send the same document or image separately for every question.

For example, an insurance application could submit a claim document, photographs, and supporting information. It could then ask several questions:

  • Is the claim document complete?
  • Is supporting evidence missing?
  • Does the damage match the submitted description?
  • Which department should review the claim?
  • What level of priority should the application assign?

The API evaluates these questions against the shared input.

This approach could simplify applications that need to make multiple decisions from the same information. However, Google specifically reports relatively stable latency from one to 100 questions under its tested conditions, not a universal latency guarantee for all 1,000 questions.

2. It Supports Five Types of Input

Gemini Decision 1 supports multimodal input, allowing developers to provide different types of information.

According to Google’s announcement, the API supports:

  • Text
  • Images
  • Audio
  • Video
  • PDF documents

Google states that the API supports shared inputs of up to 256,000 tokens.

This capability is useful for applications that process information in different formats.

For example, a business application could analyze a customer conversation recording alongside account information. A document-processing application could evaluate a PDF together with supporting photographs.

However, supporting multiple input formats does not mean the model will interpret every document or recording correctly. Developers still need to test accuracy for their specific applications.

3. It Offers Three Structured Decision Types

Gemini Decision 1 supports three main types of decisions.

Yes or No Decisions

The first option evaluates whether a particular condition is true or false.

For example, a developer could ask whether a submitted document contains a valid signature.

The model estimates the probability that the condition is true.

Multiple-Choice Decisions

The second option selects the most suitable answer from predefined choices.

Google supports between two and 255 options for this decision type.

For example, an application could determine whether a customer request belongs to billing, technical support, or account management.

Rubric Scoring

The third option evaluates information against an ordered scoring rubric.

Developers can define different levels and ask the model to assign a score based on the available evidence.

This could help applications evaluate documents, categorize content, or assess whether information meets predefined requirements.

These structured outputs make the API particularly relevant to developers building automated workflows.

4. How Is Gemini Decision 1 Different From Traditional AI Models?

The main difference is the type of output each system is designed to produce.

FeatureTraditional generative AIGemini Decision 1
Primary purposeGenerate responses and contentEvaluate predefined decisions
OutputGenerated textStructured answers and scores
Question typesOpen-ended questionsYes/no, multiple choice, rubric
Processing approachToken-by-token generationSingle forward-pass scoring
Best-suited tasksWriting, explanation, reasoningClassification and decision workflows
Confidence informationDepends on the model and interfaceIncluded in structured results

Gemini Decision 1 is not intended to replace every general-purpose AI model. Instead, it provides a specialized option for developers who need structured decisions rather than detailed written responses.

This is part of a wider shift toward specialized AI applications. For another example of AI being applied to demanding technical tasks, read our article on Anthropic Claude verification and its key AI breakthroughs.

That article examines how AI can support software verification, showing how different AI systems can be applied to specific technical problems.

5. Google Reports 150–250 Milliseconds of End-to-End Latency

Speed is another important part of Google’s announcement.

Google reports that Gemini Decision 1 delivers 150–250 milliseconds of end-to-end latency. The company separately reports 50–100 milliseconds of median model-compute time.

These figures are Google’s published measurements, not independent performance results.

Google says its latency remained nearly flat when requests increased from one to 100 questions, under the tested conditions involving contexts up to 16K tokens.

These measurements should not be interpreted as a guarantee for every input size or the maximum 1,000-question request.

Fast responses could make the API useful for applications that need to evaluate information without waiting for a lengthy generative response.

However, actual application performance will depend on network conditions, input size, integration design, and other system components.

6. Google Prices Gemini Decision 1 at $0.10 per Million Input Tokens

Google has introduced pricing intended to make the Decisions API suitable for applications that process large numbers of structured questions.

The published pricing is:

Pricing componentCost
Input tokens$0.10 per million
Output tokens$0
No-cost tierAvailable for prototyping

Google says shared input is billed once per request, regardless of how many questions are evaluated. Each question still adds its own instructions and criteria tokens.

The company also claims Gemini Decision 1 is nearly five times cheaper across its shared benchmark suite than Gemini 3.5 Flash-Lite using the Interactions API.

This cost comparison is based on Google’s benchmark methodology and should not be treated as a universal cost saving for every application.

Developers should check the official pricing and API documentation before deploying the model in production.

7. Google Is Already Using the API in Its Products

Google says several of its products are already using the Decisions API for specific tasks.

Gemini Notebook: Google says the API helps route incoming queries to an appropriate model tier and reasoning level. The company reports that this can reduce latency for everyday queries by up to 40% in that workflow.

Stitch: Google’s design tool uses the API in a rapid canvas experiment to support quick product-flow generation and editing.

CC from Google Labs: The company says the API helps select suitable emoji reactions from a palette of more than 200 options.

Google also reports using the Decisions API in Search as a safety layer for detecting adversarial behavior and content involving its tools.

These examples demonstrate how a specialized decision model can support particular functions inside larger AI applications. These product examples and performance figures are based on Google’s own announcement.

Who Can Benefit From Gemini Decision 1?

The API could be useful for several types of applications.

Software Developers

Developers can integrate structured AI decisions into applications without building every classification workflow from scratch.

Customer Support Platforms

Support systems could use the model to categorize customer requests, identify missing information, and route cases to appropriate departments.

Document Processing Companies

Businesses that process forms, reports, and applications could use structured questions to evaluate submitted documents.

AI Agent Developers

Developers building AI agents could use the Decisions API for model selection, tool routing, and other bounded decisions.

Business Automation Teams

Organizations could evaluate information against predefined rules and send uncertain cases for human review.

These are potential applications based on the API’s documented capabilities, not a guarantee that it will perform accurately in every business environment.

AI is also changing how people access and process online information. Our article on AI search in 2026 explores this broader development and its implications for search and information discovery.

Limitations Developers Should Understand

Despite its capabilities, Gemini Decision 1 has limitations.

It is not a general-purpose chatbot. Its main purpose is to evaluate predefined questions and options, not generate detailed articles, conversations, or explanations.

Results require validation. A confidence score does not guarantee correctness. Developers should test the model against representative examples before using its decisions in important workflows.

Question design matters. Poorly defined options or unclear scoring criteria can produce unsuitable results.

Application performance varies. Google’s published latency figures should not be assumed for every integration or input size.

Human review remains important. High-impact decisions should include appropriate safeguards and escalation procedures.

Understanding these limitations is essential before using the model in production systems.

Conclusion

Google’s Gemini Decision 1 introduces a specialized approach to AI-powered decision-making.

Its ability to evaluate up to 1,000 questions against shared multimodal input could help developers build classification, routing, and automation workflows.

The model is not a replacement for general-purpose AI assistants. Instead, it addresses a different requirement: making structured decisions from available information.

For developers building applications that need fast, predefined decisions, the Gemini Decisions API is worth evaluating. Its real value will depend on how accurately it performs on their own data and whether its cost and speed advantages hold in production.

Similar Posts