CryptoSlate
Google launched its Gemini Decisions API on Nov. 11, 2026, letting software request a bounded judgment instead of a written answer. Powered by Gemini Decision 1, the endpoint returns fixed choices, scores against predefined rating scales, and a confidence estimate for each answer, all usable directly in code to route requests, check policies, or score content. Google reports end-to-end latency of 150 to 250 milliseconds for contexts up to 16K tokens, with latency nearly flat across 1 to 100 questions per request.
Why it matters
The practical shift is in the response itself: a value the software can branch on, with no chatbot prose to parse. A single request can combine text, images, audio, video, and PDFs with up to 1,000 independent questions, under a 262,144-token context limit. Pricing is $0.10 per million input tokens with no output-token charge, and shared input is billed once per request.
Google says Gemini Notebook already uses the service to pick model tiers and thinking levels, cutting latency up to 40% for everyday queries, while Search applies it to detect adversarial behavior as a safety layer. Explanations, tool calls, and multi-step reasoning stay on the Interactions API.
Market impact
The caveat sits in the confidence layer. The model distributes probability among supplied options even when none fits, and user-supplied text can argue for its own classification. Google recommends calibrating thresholds on labeled traffic, adversarial tests, and a human or second check for high-stakes actions such as payouts.
For developers routing AI workloads, the flat latency curve and zero output-token billing make classification-heavy pipelines materially cheaper than generating answers first and parsing them after. Watch adoption among inference platforms, where OpenRouter already lists matching pricing.
Frequently asked questions
-
What does Google's Gemini Decisions API do differently from a normal Gemini request?
It returns fixed choices, rating-scale scores, and confidence estimates that code can branch on directly, without generating a written answer. Explanations and multi-step reasoning remain on the Interactions API.
-
How fast is the Gemini Decisions API?
Google reports end-to-end latency of 150 to 250 milliseconds for contexts up to 16K tokens, with latency nearly flat across 1 to 100 questions per request.
-
How much does the Gemini Decisions API cost?
Google lists $0.10 per million input tokens with no output-token charge. Shared input is billed once within each request, while each question adds its own instruction and criteria tokens.
-
What are the limits of a Decisions API request?
A request can combine text, images, audio, video, and PDFs with up to 1,000 independent questions, under a total context limit of 262,144 tokens. Questions cannot see one another's answers, so code must combine the results.
-
Why does Google recommend human checks on Decisions API outputs?
The model distributes probability among supplied options even when none fits, and user-supplied text can argue for its own classification. Google advises calibrating confidence thresholds on labeled traffic and adding a human or second check for high-stakes actions such as payouts.
WatcherGuru
CoinTelegraph
Lookonchain
Crypto News
WuBlockchain
CoinDesk