Many MindedGoogle releases a cheap, fast model built only to make closed-ended calls
Google Deepmind has launched the Gemini Decisions API, powered by a model called Gemini Decision 1. It is built for closed-ended judgments: yes/no questions, multiple choice and scale-based ratings. It is available now through Google AI Studio, with integrations for OpenRouter, LangChain, Vercel and other platforms rolling out.
The model takes text, images, audio, video and PDFs, and can handle up to 1,000 independent questions in a single pass. Instead of writing out an answer, it scores the options directly and returns probabilities with confidence values. Google reports response times of 150 to 250 milliseconds. In its example, a photo of a kitchen counter produced a stain yes/no at 0.95 confidence, a cleaning action chosen at 98 percent, and a severity rating of 1.19 on a 0-2 scale.
Pricing is $0.10 per million input tokens, with no charge for output tokens. Google says that is about five times cheaper than Gemini 3.5 Flash-Lite. It says it already uses the API internally for model routing in Gemini Notebook, design prototyping in Stitch, and safety checks in Google Search.
Why it matters: much of the work of sorting, filtering and triaging information is a long series of small judgments. Pricing those judgments at a fraction of a cent, across every media type, lowers the cost of applying machine attention to large piles of material. That could help with things like moderation, document review and routing questions to the right tool, for developers who could not previously afford it.
Caveats: the benchmark chart, which shows leads in agent routing, reasoning, multimodal understanding and calibration error, and the five-times price comparison are Google's own claims, not independently verified. So are the internal usage claims. Jev, OpenAI, Cloudflare and Microsoft have all released similar decision models in recent weeks, so independent comparisons should follow.
Source: The Decoder