Google has released the Gemini Decisions API, a developer endpoint that answers fixed-choice questions instead of writing text, joining other large AI companies in selling a model built for the small judgments inside AI agents. The service runs on a new model, Gemini Decision 1, and costs $0.10 per million input tokens with no charge for output, according to a Nov. 11 Google blog post.
Google describes the endpoint as "purpose-built primitives for bounded choices like tool selection, policy checks, model routing, and context grading." One request can pair a shared input of up to 256,000 tokens, in text, images, audio, video or PDFs, with up to 1,000 independent questions.
Scoring options instead of writing answers
Rather than generating an answer token by token, the model scores candidate options in a single forward pass and returns probabilities, Google said. The company puts end-to-end latency at 150 to 250 milliseconds for inputs up to 16,000 tokens, nearly flat from one to 100 questions per request.
Developers can ask yes-or-no questions, which return the probability that a condition holds, or multiple-choice questions with 2 to 255 options. A separate rubric type grades input against an ordered scale. Every answer carries a confidence score from 0 to 1, which Google says lets code send ambiguous cases to a fallback or a human reviewer.
The shared input is billed once per request, according to Google, so each added question costs only the tokens in its own instructions and criteria. VentureBeat worked through Google's pricing: a 4,000-token record checked against 20 questions of 50 tokens each costs about $0.50 per 1,000 records in one request, versus $8.10 per 1,000 records if each check resends the record in a separate call.
Google's own uses
Teams inside Google already use the API, the company said. Gemini Notebook uses it to route queries to different model tiers and thinking levels, serving everyday queries with up to 40% lower latency, according to the blog post. Google Search uses it as a safety layer for MCP tools to detect adversarial behavior, the company said.
The API is live on the Gemini Developer API's v1beta endpoint with a no-cost tier in Google AI Studio. Google said integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM and other frameworks, with support in Google Cloud coming soon. Its documentation says the endpoint does not support context caching or the Batch API, so input repeated across requests is billed in full.
A crowded category
Google is not first. VentureBeat reported that OpenAI made its own Decisions API, running on GPT-6 Luna, generally available on Oct. 22 at the same $0.10 base rate, with cached input reads at $0.01 per million tokens.
Amazon released Strands Decider 2B in October, VentureBeat reported, as an open model under the Apache 2.0 license that teams can run on their own hardware. Startup TypeSafe prices its Jev 1.14 model at $0.042 per million input tokens, according to VentureBeat.
In Google's launch benchmarks, Gemini Decision 1 led five of seven tests that all four compared models ran, scoring 92.2% on JevBench against 90.9% for GPT-6 Luna and 90.0% for Jev 1.14, VentureBeat reported. Google said its model had the lowest calibration error in its comparison, 0.0250, roughly half of Jev's 0.0494.
Independent results were less favorable. The Artificial Analysis Decision Index v2.1 ranked Gemini Decision 1 ninth of 18 models at launch with a score of 65.98, behind Nace AI's Drex Max v1.1 at 72.62 and Jev 1.14 at 66.23, according to VentureBeat. On a test that asks 15 questions about each document, Gemini answered every question correctly on 33% of documents, compared with 63% for Drex Max.
VentureBeat also cited a prompt-injection test in which a single planted field dropped Jev's probability of blocking an action from 0.76 to 0.48. The publication advised developers to keep approval requirements and transaction limits in application code rather than leave them to a decision model.
Source: Google, “Gemini Decisions API: Fast, calibrated multimodal decisions”
Additional sources: Google AI Studio docs, “Decisions API”; VentureBeat, “Google's Gemini Decision 1 model scores up to 1,000 questions in one API call”; MarkTechPost, “Google Launches Gemini Decisions API With Gemini Decision 1”
