Released by Google on November 11: the new model Gemini Decision 1 plus the Decisions API, built for closed-form decisions (tool selection, policy checks, model routing, context scoring). A single request can grade 1,000 yes/no, multiple-choice or scoring questions on the same multimodal input (text/image/audio/video/PDF, up to 256K tokens), returning probabilities and confidence in a single forward pass, with end-to-end latency of 150–250ms, $0.10 per million input tokens, output free, and a free tier in AI Studio. HN 416 points / 125 comments, Pichai posted on X, and Latent Space called November 11 “decision model day” (OpenAI had a similar API in September, and TypeSafe Jev came earlier).
Suggested action:Worth trying; directly usable for agent routing / tool-selection scenarios.
| date | heat | rank | title |
|---|---|---|---|
| 2026-11-12 evening | 131 | 5 | Gemini Decisions API (Gemini Decision 1) |