Google Launches Gemini Decisions API to Handle 1000 Questions in One Request

Google Enters Decision Model Market with Gemini Decisions API

Google has launched the Gemini Decisions API, a specialized service designed to handle routine decision-making tasks without generating text responses. Powered by the new Gemini Decision 1 model, the API allows developers to evaluate up to 1,000 questions in a single request against shared multimodal inputs including text, images, audio, video, and PDFs.1

1

2

2

The service arrives weeks after OpenAI launched its competing Decisions API, marking Google's entry into the rapidly growing decision model market.1

1

Unlike conventional generative AI models that produce responses token by token, Gemini Decision 1 scores predefined options in a single forward pass, returning probabilities and confidence scores with remarkably low latency. Google DeepMind reports end-to-end latency of 150-250 milliseconds in tests involving contexts of up to 16,000 tokens, with performance remaining nearly flat from 1 to 100 questions per request.1

1

4

4

The median model computation time ranges from 50-100 milliseconds, demonstrating the efficiency gains from closed-set decision-making.2

2

Cost Efficiency and Pricing Structure

Google has priced the Gemini Decisions API at $0.10 per million input tokens, with zero output-token charges. This pricing model offers significant cost advantages for applications that need to evaluate multiple questions against the same data. The shared input is billed once per request regardless of how many questions are evaluated, though each question adds its own instruction and criteria tokens.1

1

4

4

According to Google, the Gemini Decision 1 model is nearly five times cheaper than Gemini 3.5 Flash-Lite using the Interactions API across its shared benchmark suite.2

2

5

5

In a practical example, processing a 4,000-token record with 20 checks that add 50 tokens each would cost approximately $0.50 per 1,000 records. Making 20 separate calls that resend the record would cost $8.10 per 1,000 records, demonstrating substantial savings.3

3

Three Decision Types with Multimodal Support

The Gemini Decisions API supports three distinct question types through purpose-built primitives for bounded choices. Developers can use yes/no checks to estimate the probability that a condition holds, multiple-choice questions to select from 2 to 255 named options with normalized probabilities for each candidate, or rubric-based scoring to grade inputs across an ordered scale starting at 0.4

4

3

3

Built on Gemini's native multimodal architecture, the API can process text, images, audio, video, and PDFs directly in a single forward pass. The service supports a context window of up to 256,000 tokens, allowing developers to combine multiple data types simultaneously without relying on separate processing systems.2

2

4

4

Developers can also adjust image and video resolution to control visual detail and token cost, providing flexibility in managing resource consumption.3

3

Competition with OpenAI and Market Positioning

The launch positions Google directly against OpenAI's Decisions API, which became generally available in October and supports up to 400 questions per request compared to Google's limit of 1,000.1

1

Both services charge $0.10 per million input tokens at standard rates with no output-token charges, though OpenAI offers discounted pricing for cached inputs.1

1

Google's offering distinguishes itself through broader multimodal support, including audio and video inputs that OpenAI's service lacks.1

1

In benchmark testing, Google reports that Gemini Decision 1 outperformed competing systems across seven of nine tests, including achieving 92.2% on JevBench compared to GPT-6 Luna's 90.9% and Jev 1.14's 90.0%.2

2

3

3

The model also demonstrated the lowest calibration error at 0.0250, roughly half of Jev's 0.0494, indicating more reliable confidence scores.3

3

Real-World Applications and Use Cases

Google is already deploying the technology across several of its products, demonstrating practical applications for AI decision-making. In Gemini Notebook, the API routes incoming queries to appropriate model tiers and reasoning levels, reducing latency for everyday queries by up to 40% while reserving more powerful reasoning models for complex tasks.1

1

2

2

Google Search uses the technology as a safety layer to detect adversarial behavior involving Model Context Protocol (MCP) tools, extending its use beyond model routing to security-related decisions.1

1

4

4

The technology also powers rapid design experiments in Google's Stitch platform, enabling users to generate interactive product wireframes and make changes through voice commands or touch inputs. Google's CC assistant leverages the API to select emoji reactions from more than 200 options in a fraction of a second.2

2

Lian Jye Su, chief analyst at Omdia, identified policy checks, risk assessment, and credit scoring as potential enterprise applications. "The model is great at generating decisions from unstructured, dynamic input," Su noted.1

1

A claims intake system, for example, could send a scanned form, photos of damage, and notes from a customer call in one request, asking which queue should receive the claim, whether evidence is missing, and how urgently someone should review it.3

3

Enterprise Considerations and Risk Factors

While the API offers significant advantages, experts caution about potential risks in production environments. Su warned that changes in incoming data or deliberately manipulated inputs could cause decision-making APIs to produce incorrect judgments. "All decision-making processes must be highly contextual with support from a trusted and secure context layer," Su emphasized.1

1

He recommended connecting decision models to databases containing historical transactions and established business policies to ensure models evaluate requests against relevant organizational information.

Tulika Sheel, senior vice president at Kadence International, raised concerns about human accountability as enterprises incorporate AI decision-making into autonomous workflows. "As AI agents become more autonomous, enterprises will need clear accountability for decisions and actions taken on behalf of users," Sheel stated.1

1

Su also noted that Google would need to meet enterprise requirements for data retention and sovereignty without undermining the cost and performance advantages of the service.1

1

Availability and Integration Options

The Gemini Decisions API is now available through Google AI Studio, which includes a free tier for prototyping with limits of 200 requests and 50,000 input tokens per minute.3

3

Ecosystem integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, and LiveKit alongside the google-genai SDKs.4

4

Support for Gemini in Google Cloud is expected later, though Google has not specified a release date.1

1

For teams testing with customer records, the Gemini API terms distinguish between paid and unpaid access. Google excludes paid-service prompts and responses from product improvement and retains logs for a limited period for policy enforcement and required disclosures, while unpaid-service terms allow product improvement uses with regional exceptions.3

3