Why Google and OpenAI want AI agents to decide faster

Google on November 11 introduced the Gemini Decisions API, a developer tool designed to help AI agents make quicker choices as they carry out tasks, joining AI startups OpenAI and TypeSafe in an emerging market for AI that specialises in decisions.

Powered by Gemini Decision 1, the API can help software select tools, route requests to different AI models, check content against policies and assess information, Google said.

TypeSafe introduced its Jev model on September 15, while OpenAI made its Decisions API generally available on October 22.


Why small decisions matter

An AI agent may need to pick a tool, classify a request or decide whether to seek help before completing a task. General-purpose large language models can handle these choices, but generating an answer token by token adds time and cost, especially when an agent repeats the process across several steps.

Decision models focus on questions with defined answers. They score the available options in a single processing pass and return probabilities or scores that software can use directly, without generating a written response.

For example, an agent could assess whether a request needs a web search, a calculator or a database lookup, then use the result to choose a tool. Developers can also use confidence scores to send uncertain cases to a human or another model.

How the offerings compare

Google highlights its ability to process text, images, audio, video and PDFs together. The API can evaluate up to 1,000 independent questions against a shared input within a 256K-token context window, according to the company.

It supports yes/no checks, multiple-choice questions with up to 255 options, and scoring against a defined set of criteria.

OpenAI’s API, powered by GPT-6 Luna, supports text, images and structured data, and handles up to 400 questions per request. It also integrates with GPT-Live for voice applications, the company said.

TypeSafe describes Jev as a “System One” model, built with a different architecture and training method for structured, probabilistic judgements. It says application code should combine those judgements with business rules to determine an action.

Pricing and early use

Google charges $0.10 per million input tokens, the same base rate OpenAI lists for its Decisions API. TypeSafe charges $0.042 per million input tokens for Jev. All three list zero output-token charges.

Google said its API costs about a fifth as much as Gemini 3.5 Flash-Lite through the Interactions API across its benchmark suite.

The company said Gemini Notebook uses the API to select models and reasoning levels, cutting response times by up to 40% for everyday queries. Its design tool Stitch and Google Labs’ CC also use the API to speed up interactions.

Developers can access the API through Google AI Studio, while integrations with platforms including LangChain, OpenRouter, Vercel and LiteLLM are rolling out, Google said.