Google Launches Gemini Decision 1, an AI Model That Scores 1,000 Questions in a Single Pass

Google has entered the growing market for AI decision-making models with the launch of Gemini Decision 1, a specialised artificial intelligence model designed to evaluate up to 1,000 questions in a single processing pass. Developed by Google DeepMind, the model is built for fast, structured decisions rather than generating long conversational responses. It can analyse text, images, audio, video and PDF documents together, returning answers with probability and confidence scores.

Announced on November 11, 2026, Gemini Decision 1 is available through Google’s Gemini Decisions API in Google AI Studio. The company says the system can deliver end-to-end response times of approximately 150 to 250 milliseconds, with pricing of $0.10 per million input tokens and no charges for output tokens. The launch positions Google in an emerging category of AI models designed to make thousands of small decisions quickly and economically, particularly for businesses operating automated workflows and AI agents. <Cite refs={[“turn738359search0″,”turn738359search1″,”turn738359search4”]}/>

What Is Google Gemini Decision 1?

Gemini Decision 1 is a specialised AI model developed for what Google calls closed-set decisions.

A conventional generative AI model receives a prompt and produces a response, usually generating text one token at a time. This approach is useful for writing, coding, summarisation and complex reasoning, but it can be unnecessarily expensive or slow when software only needs to select an option or evaluate a condition.

Gemini Decision 1 takes a different approach. Instead of generating a lengthy response, it scores predefined options and returns probabilities indicating which choices are most likely to be correct.

Developers can submit one shared input alongside as many as 1,000 independent questions. The model evaluates those questions in a single forward pass and returns the results with confidence scores.

For example, an insurance company could submit a claim document, photographs of damaged property and a customer call recording. It could then ask the model whether required documents are missing, which department should handle the claim and how urgently the case needs human attention.

This makes the model particularly suitable for applications where software needs to make numerous structured decisions based on the same information. <Cite refs={[“turn738359search0″,”turn738359search4”]}/>

How Can Gemini Decision 1 Evaluate 1,000 Questions at Once?

The main innovation behind Gemini Decision 1 is its ability to evaluate multiple questions against shared information without repeatedly processing the same input through separate requests.

Traditional AI workflows often require developers to send multiple prompts to a model, particularly when a document needs to be checked against several conditions.

For example, a company reviewing an invoice might need to verify the supplier, check the payment amount, identify missing information, determine whether approval is required and classify the expense.

With conventional workflows, each check could involve a separate model request or a more complicated prompt.

Gemini Decision 1 allows developers to send the document once and define multiple questions against it.

The system supports up to 1,000 independent questions in one request, while the shared input can contain up to 256,000 tokens.

Google says the model’s shared input is billed only once, regardless of how many questions are evaluated in the request. Each question adds only its own instructions and evaluation criteria to the billed input.

This approach could reduce repeated processing, lower API costs and make large-scale automated decision systems more practical. <Cite refs={[“turn738359search0″,”turn738359search4”]}/>

Three Types of Decisions Supported by Gemini Decision 1

Google’s Decisions API supports three primary decision formats.

Decision type	How it works	Example
Yes-or-no	Estimates the probability that a condition is true	Does this document contain a valid signature?
Multiple choice	Selects the most suitable option from a list	Which department should receive this request?
Rubric scoring	Evaluates information against an ordered scoring scale	How severe is the damage shown in these photographs?

1. Yes-or-No Decisions

Developers can ask the model whether a particular condition is satisfied.

For example, an e-commerce platform could check whether a product image contains prohibited material or whether a customer support request includes enough information to proceed.

The model returns a probability associated with the answer.

2. Multiple-Choice Decisions

The API can select from between two and 255 predefined options.

This could help businesses classify customer requests, select the appropriate AI model or route transactions to the correct operational team.

3. Rubric-Based Scoring

The model can also evaluate information against an ordered set of criteria.

A company could use this feature to assess document completeness, classify the severity of an incident or evaluate the quality of a submitted application.

Each decision includes a confidence score between zero and one. Developers can use these scores to determine whether an automated action is appropriate or whether the case should be reviewed by a human. <Cite refs={[“turn738359search0”]}/>

Gemini Decision 1 Supports Text, Images, Audio, Video and PDFs

Another important feature is the model’s multimodal capability.

Gemini Decision 1 can evaluate different types of information within the same request rather than requiring developers to convert everything into text first.

Supported formats include:

• Text: Written documents, instructions and business records.
• Images: Photographs, scanned documents and visual evidence.
• Audio: Recordings of customer calls and other spoken information.
• Video: Visual recordings that need to be assessed against specific criteria.
• PDFs: Contracts, invoices, reports and other business documents.

The ability to combine these formats could be particularly useful for insurance, healthcare administration, financial services and customer support.

For example, an insurance workflow could analyse a scanned claim form alongside photographs of vehicle damage and an audio recording from the customer.

The system could then answer multiple questions about the evidence, identify missing information and recommend an appropriate processing category.

Google says the API can handle multimodal inputs of up to 256,000 tokens, allowing developers to evaluate substantial amounts of information in a single request. <Cite refs={[“turn738359search0″,”turn738359search4”]}/>

Google Claims 150–250 Millisecond Response Times

Speed is one of the central selling points of Gemini Decision 1.

Google reports end-to-end response times of approximately 150 to 250 milliseconds under its stated testing conditions. The model’s median processing time is approximately 50 to 100 milliseconds.

Unlike generative models that produce responses token by token, the decision model scores predefined options directly.

This makes it suitable for applications where decisions must happen quickly, such as routing incoming requests, selecting AI models and checking whether content violates a policy.

Google also says the model’s latency remains relatively stable when the number of questions increases from one to 100, under the benchmark conditions it tested.

However, actual performance can vary depending on input size, media processing, network conditions and the complexity of the questions. Developers handling the maximum 1,000 questions should test their own workloads rather than assume every request will achieve the same response time. <Cite refs={[“turn738359search0″,”turn738359search4”]}/>

Gemini Decision 1 Pricing: Nearly Five Times Cheaper Than Flash-Lite in Google’s Comparison

Google has priced Gemini Decision 1 at $0.10 per million input tokens, with no charges for output tokens.

The company says the shared input is billed only once per request, even when developers evaluate multiple questions against the same material.

This pricing structure could make a difference for businesses processing large volumes of documents or customer interactions.

Google also compared the Decisions API with Gemini 3.5 Flash-Lite using a shared benchmark suite. It reported that Gemini Decision 1 was nearly five times cheaper for that particular workload.

Pricing metric	Gemini Decision 1
Input token price	$0.10 per million tokens
Output token price	$0
Shared input billing	Once per request
Questions supported	Up to 1,000
End-to-end latency	150–250 milliseconds

The company has also made a no-cost tier available through Google AI Studio for developers who want to experiment with the API.

The comparison with Flash-Lite is based on Google’s benchmark workload and should not be treated as a universal cost advantage for every application. <Cite refs={[“turn738359search0″,”turn738359search4”]}/>

Google Is Already Using the Model Across Its Products

Google says several of its internal teams are already using the Decisions API.

One application is Gemini Notebook, where the model helps route incoming queries to the appropriate AI model and reasoning level.

The objective is to send straightforward requests to faster, less expensive models while reserving more advanced reasoning systems for complicated questions. Google says this approach can reduce latency for routine queries by up to 40% in that workflow.

The company is also using the model in Stitch, its AI-powered product design environment, to support faster design iteration and interaction.

Another application is Google Search, where the Decisions API acts as a safety layer for identifying adversarial behaviour and potentially harmful content involving tools.

Google’s CC application also uses the model to select appropriate emoji reactions from a large set of options.

These examples demonstrate how decision models can work behind the scenes, making products faster without requiring users to interact with a separate AI assistant. <Cite refs={[“turn738359search0″,”turn738359search1”]}/>

Google Enters a Growing Market for AI Decision Models

Google’s announcement comes as technology companies increasingly explore AI systems designed specifically for classification, routing and structured decision-making.

These models are different from general-purpose chatbots because their primary purpose is not to generate lengthy responses. Instead, they evaluate available options and return structured results that other software can use.

Startups and established technology companies are developing similar approaches.

The Decoder reported that companies including Jev, OpenAI, Cloudflare and Microsoft have also entered this emerging category.

This market is particularly relevant to agentic AI. AI agents often need to make frequent decisions about which tools to use, which information is relevant and what action should happen next.

Using a large generative model for every small decision can increase latency and operating costs.

A specialised decision model could handle routine evaluations while more powerful models focus on complex reasoning, planning and content generation.

However, performance still depends on the specific task. Independent benchmark results show that different models can lead in different categories, meaning developers should compare accuracy, reliability and pricing before selecting a system. <Cite refs={[“turn738359search1″,”turn738359search4”]}/>

The Bigger Picture

Gemini Decision 1 highlights an important shift in the AI industry: companies are beginning to build specialised models for individual stages of automated workflows rather than relying on one general-purpose model for every task. Generative models remain valuable for writing, coding and complex reasoning, but many business operations require simpler decisions that must happen quickly and repeatedly. A model capable of evaluating hundreds of conditions against shared information could make large-scale automation more efficient.

The development could also strengthen the infrastructure behind agentic AI. As businesses deploy more AI agents, they will need systems that can route requests, check policies, assess information and select appropriate actions without repeatedly invoking expensive generative models. Google’s launch adds another option to this emerging market, although real-world adoption will depend on decision accuracy, confidence calibration, integration costs and how effectively companies manage cases that require human judgement.

Looking Ahead

Google’s immediate focus will be expanding developer access and integrations for Gemini Decision 1 through its API ecosystem. Developers can already experiment with the model through Google AI Studio, while integrations involving platforms such as OpenRouter, LangChain, Vercel, LiteLLM and Pydantic AI are being rolled out. Support for Gemini in Google Cloud is also expected to follow.

The broader competition will depend on which decision models can deliver the best combination of speed, accuracy, reliability and cost. For businesses building AI-powered operations, the ability to evaluate thousands of conditions quickly could become an important part of their technology infrastructure. Gemini Decision 1 gives Google a new position in that market as the industry moves towards more specialised and efficient AI systems.

---

SEO Metadata

SEO Title: Google Launches Gemini Decision 1 AI Model That Scores 1,000 Questions

Meta Description: Google introduces Gemini Decision 1, a multimodal AI model that evaluates up to 1,000 questions in one pass with 150–250 ms response times.

URL Slug: google-gemini-decision-1-1000-questions

Focus Keyphrase: Google Gemini Decision 1

Secondary Keywords: Gemini Decisions API, Google AI model, AI decision model, multimodal AI, agentic AI, Google DeepMind, AI automation, AI model routing, Gemini API pricing