Google unveiled the Gemini Decisions API on Wednesday, a specialized system built to handle the high-volume classification and routing decisions that AI applications perform repeatedly throughout the day. Unlike traditional generative models that compose text responses, this approach prioritizes speed and cost efficiency by evaluating a fixed set of options in a single inference pass.

The underlying model, Gemini Decision 1, delivers response times between 150 and 250 milliseconds across tested workloads. Pricing begins at $0.10 per million input tokens, with no separate charge for output tokens. The announcement follows OpenAI's recent Decisions API launch and reflects growing industry interest in models designed for bounded-choice scenarios rather than open-ended text generation.

Single-pass evaluation without text generation

The API manages a 256,000-token context window and can process up to 1,000 independent questions against a single input, including multimodal data. Each response includes probability scores and confidence metrics that let developers select among predefined options, validate information against defined criteria, and determine application behavior. Google CEO Sundar Pichai characterized the model on X as "built for closed-set choices," while Google DeepMind product manager Min Choi and principal scientist Felix Yu called it a provider of "purpose-built primitives for bounded choices" in the official announcement.

Internal deployment and use cases

Google has already integrated Gemini Decision 1 into its own systems. Within Gemini Notebook, real-time model and reasoning-level routing has cut latency by as much as 40%. The model also powers adversarial-behavior detection in Google Search related to MCP tools.

Benchmark results and competitive positioning

Google's internal testing shows Gemini Decision 1 outperforming OpenAI's GPT-6 Luna, though the gap narrows when compared to TypeSafe's Jev 1.14. On JevBench, Gemini Decision 1 scored 92.2%, Luna achieved 90.9%, and Jev reached 90%. Logan Kilpatrick, who leads product for Google AI Studio and the Gemini API, described it on X as "our new frontier decision model in the Gemini API."

Jev outperformed Google's model on two of seven shared benchmark tests and demonstrated lower operational costs, consuming $0.68 for the complete benchmark suite versus $0.88 for Gemini Decision 1. Google's model recorded the best calibration error among the tested systems. However, all benchmark data originates from Google's own testing; independent evaluations have not yet compared these models under production conditions. SiliconANGLE also noted that Jev offers lower per-token pricing than Google's offering.

All of the benchmark figures come from Google's own testing, and independent evaluations have yet to establish how the models compare in production.

Pricing and availability

The pricing model charges developers once for sending shared context, regardless of how many questions they ask about it, though the questions themselves consume tokens. Each question receives independent evaluation, meaning workflows that depend on one answer to formulate the next question require an additional API call.

Gemini Decision 1 is accessible through Google AI Studio, which offers a free tier for prototyping. Integrations are available via OpenRouter, LangChain, Vercel, and LiteLLM. Support through Google Cloud remains in development, with no announced launch timeline.

Developers only pay to send the shared context once, even when they're asking multiple questions about it…