Google Launches Gemini Decisions for 1,000 Questions at Once

Google DeepMind commercializes Gemini Decision 1, a model designed to handle closed decisions in a single pass across multiple formats. The company highlights benchmark results and a very low usage fee, while other players have just entered this niche.

A Crowded Niche and Claimed Performance

Comparable decision models have been announced in recent weeks. Following a version unveiled by the startup Jev, OpenAI, Cloudflare, and Microsoft quickly launched similar offerings. Google DeepMind claims that Gemini Decision 1 tops a suite of benchmarks for agent routing, reasoning, and multimodal understanding, achieving the lowest Top-1 ECE calibration error.

Making Decisions in a Single Pass on Text, Image, Audio, Video, and PDF

Gemini Decision 1 is capable of processing up to 1,000 distinct questions simultaneously and supports inputs in the form of text, images, audio, videos, or PDF files. The API is aimed at closed decisions such as yes/no answers, multiple-choice selections, or ratings on a scale, providing probabilities accompanied by confidence levels within a timeframe of 150 to 250 milliseconds. An example provided illustrates three tasks on a counter photo: detecting a stain with 95% confidence, selecting a cleaning action at 98%, and estimating severity at 1.19 out of 2.

Immediate Availability, Ongoing Integrations, and Low Pricing

The API is available now via Google AI Studio, with integrations planned for OpenRouter, LangChain, Vercel, and other platforms currently being deployed. The announced pricing is $0.10 per million input tokens, with no charges for output tokens, a level that Google indicates is about five times lower than Gemini 3.5 Flash-Lite. According to Google, the API is already being used internally for dynamic model routing in Gemini Notebook, design prototype creation in Stitch, and security checks in Google Search.