Gemini Decisions API
Points and comments are a snapshot, not live.
Google introduces a fast, calibrated multimodal decision API paying zero output tokens.
Google DeepMind launched the Gemini Decisions API, powered by the Gemini Decision 1 model. It can evaluate up to 1,000 yes/no, choice, and rubric-scoring questions in a single forward pass across text, images, audio, video, and PDFs. Latency is 150-250 ms end-to-end at $0.10 per 1M input tokens with no output token charges. Google teams use it for real-time model routing, rapid design iteration, emoji selection, and safety classification. Gemini Decision 1 leads benchmarks for calibration error and is available through the Gemini API in Google AI Studio.
What commenters are saying
Commenters compare the API to Jev and OpenAI's decision endpoint, noting Gemini's advantage in multimodal input (audio, video, PDFs).
Some note higher cost than other decision models but acknowledge the value for specific use cases. Several express skepticism about Google's track record, citing the killing of Jules, though others argue enterprise APIs differ from consumer experiments. A technical subthread discusses how model probability calibration works, contrasting it with simple logprob-based or constrained generation approaches.