Google launches Gemini Decisions API for scoring choices
Google says its new model returned results in 150-250 ms end to end in tests with contexts up to 16K tokens and was nearly 5x cheaper than a comparison using Gemini 3.5 Flash-Lite.
In this article
Sole source: Google. Google says it launched the Gemini Decisions API on 11 November 2026, powered by its new Gemini Decision 1 model. Developers can access it through the Gemini API in Google AI Studio, where they can obtain an API key; the Decisions API documentation includes Python, JavaScript and REST guides. The interface is designed for tasks such as routing queries and checking policy rules against predefined choices. Google says the model scores those choices in a single pass rather than generating answer text one token at a time. Tokens are units used to measure the input a model processes.
How the scoring works
One request can combine text, images, audio, video and PDFs in a shared input of up to 256K tokens, with up to 1,000 independent questions. Google says the model returns probabilities and confidence scores for the choices.
A yes/no question estimates the probability that a condition holds. Multiple choice selects a winner from 2 to 255 options and returns normalized probabilities for all of them. An ordered rubric grades an input using levels starting at 0 and returns a continuous weighted score and probabilities for each level. Each answer also has a confidence score from 0 to 1, which developers can use to send ambiguous cases to a fallback or a human reviewer.
Price and performance claims
Google charges $0.10 per million input tokens for Gemini Decision 1 and says there are no output-token charges. A shared input is billed once per request, while each question adds its own instruction and criteria tokens. The company also offers a no-cost tier for prototyping.
Google says median model compute time was 50-100 ms, with the end-to-end result time cited above, across different input types and contexts up to 16K tokens. It says latency stayed nearly flat in tests with 1 to 100 questions per request. The company also says the new model and API were nearly 5x cheaper across its shared benchmark suite than Gemini 3.5 Flash-Lite used with the Interactions API. These performance claims have not been independently verified.
Google claims Gemini Decision 1 leads its benchmark suite across agent routing, reasoning, ranking and multimodal understanding, with the lowest calibration error. The company says Gemini Notebook uses the API to route queries to model tiers and thinking levels, while Google Search uses it as a safety layer for MCP tools to detect adversarial behavior and content.
Google says integrations with OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode are rolling out alongside the google-genai SDKs and support for Gemini in Google Cloud is coming soon.
Earlier in Technology
SoftwareCanonical status page lists Ubuntu login and software outages
Ubuntu cloud images, Launchpad package archives, Livepatch and Landscape were also affected. The cause is unconfirmed, while email and Discourse forums remained available.
1 min
SoftwareAWS launches public-domain emulator built by agents, company says
AWS says its agents built and will maintain AWS Emulator, a tool released under a public-domain dedication for testing cloud applications locally.
4 min
InfrastructureUK announces £90 million (about $119.9 million) for Atlantic Net
The Defence Ministry selected Ocean Infinity Solutions as a commercial partner for the planned North Atlantic surveillance network, under a framework it said was due to be signed on 10 November 2026.
2 min
