ENPT

Google ships Gemini Decision 1 and the Gemini Decisions API at $0.10 per 1M input tokens

Last updated:

Source · Model card: Gemini Decision 1

On 11 Nov 2026 Google released two things together. The product is the Gemini Decisions API. The model is Gemini Decision 1 (gemini-decision-1), served on the Gemini Developer API in Google AI Studio. The blog is dated that day. The changelog lists the release under 11 Nov 2026. @googledevs posted it at 18:26 UTC. Logan Kilpatrick’s post at 18:16 UTC is the same announcement, ten minutes earlier.

The guide and the pricing page confirm the contract:

  • One shared input: text, image, audio, video and PDF, 262,144 tokens (the blog says 256K).
  • 1 to 1,000 questions per request. Types: noul (yes/no), choice (2–255), score (ordered rubric, 2–255).
  • Each answer returns probabilities and a confidence score. For choice and score, confidence equals the top probability.
  • $0.10 per 1M input tokens on the paid tier. Output tokens are 0, so the output price is “not applicable”. The free tier lists input as free of charge (200 requests/min, 50,000 tokens/min, 10,000 requests/day).
  • POST /v1beta/decisions (CreateDecision), google-genai SDK 3.2+, api_version: "v1beta".
  • The changelog says “Released”. It does not use the “generally available (GA)” label. The blog says support in Gemini on Google Cloud is still coming.

Vendor claims (UnverifiedClaim), from the blog. End-to-end latency 150–250 ms and model compute 50–100 ms. “Leads across the suite, with the lowest calibration error” (the table is a figure; the text has no ECE). Nearly 5× cheaper than Gemini 3.5 Flash-Lite on Google’s own suite. Already used in Gemini Notebook, Stitch, CC and Search. A longer integration list is “rolling out”.

Independent board. The live Artificial Analysis Decision Index page is titled v2.1. Read on 11 Nov 2026, Gemini Decision 1 scores 65.98, 10th of 19, 211 ms per task, about $345 per 1M tasks. Jev 1.14 on the same page is 66.23. This is a different board from the v2 table of 8 Nov and from the Hugging Face Decision Index. There is no Decision Index Full score, so accuracy is 6, not the 8 × Full / 59.22 line, and not 9.

Catalog page: Gemini Decision 1, scored 6.2 / 7.2 / 13. LangChain’s GoogleGenAIClassifier is in beta. OpenRouter lists the model at the same $0.10 / $0.

Same calendar day, AIE CODE was on in San Francisco. Gemini Decision 1 was not a mainstage talk in the Day 1 livestream, and Sundar Pichai and Logan Kilpatrick are not on that agenda. See AIE CODE Day 1.