Google launches Gemini Decisions API for fast yes/no and scoring calls
Gemini Decision 1 scores up to 1,000 closed questions in one pass across text, images, audio, video and PDFs, at $0.10 per million input tokens, Google says.
Key takeaways
- Google has launched the Gemini Decisions API, powered by a new Gemini Decision 1 model, for closed-set decisions instead of generated text.
- Google says one request can carry up to 1,000 yes/no, multiple-choice or rubric questions across text, images, audio, video and PDFs.
- Google puts median latency at 150 to 250 milliseconds end to end for contexts up to 16K tokens and prices input at $0.10 per million tokens; output tokens are not billed.
- Google says Gemini Notebook, Stitch, CC and Google Search already use the API.
- Similar decision models have recently come from Jev, OpenAI, Cloudflare and Microsoft, according to The Decoder.
Google has released the Gemini Decisions API, a new endpoint built for fast, closed-set decisions rather than generated text, the company said in a blog post by Google DeepMind staff on Wednesday. It runs on a new model called Gemini Decision 1 and developers can reach it via the Gemini API in Google AI Studio. Google says developers can send one input, in text, images, audio, video or PDFs, together with up to 1,000 separate questions, and get back probabilities and confidence scores for each answer.
What does the API do?
According to Google, the API is meant for bounded choices such as picking a tool, checking a policy, routing a request to a model or grading content. Instead of writing out an answer word by word, the model scores the candidate options in a single forward pass, the post says.
Google lists three question types that can be mixed in one request: yes/no checks that return the probability a condition holds; multiple choice across 2 to 255 options; and scoring against an ordered rubric. Each answer carries a confidence score between 0 and 1, which Google says lets software send unclear cases to a fallback or to a human reviewer. Inputs can be up to 256K tokens, and different media can be combined, for example a call recording with account data, according to the company.
How fast and how cheap is it?
Google says median end-to-end latency is 150 to 250 milliseconds, staying nearly flat whether a request carries one question or 100, for contexts up to 16K tokens. Pricing is $0.10 per million input tokens, with no charge for output tokens, and the shared input is billed once per request regardless of how many questions are asked, though each question adds its own instruction tokens, the post says. Google says that across its shared benchmark suite this makes the API nearly 5x cheaper than using Gemini 3.5 Flash-Lite through the Interactions API.
The company also says Gemini Decision 1 leads the benchmark suite Google reported, which covers agent routing, reasoning, ranking and multimodal understanding, with the lowest calibration error. These are Google's own benchmark results; NextFish has not independently verified them. A no-cost tier is available in Google AI Studio for prototyping, Google says.
Where is Google already using it?
Google says several of its own products use the API. Gemini Notebook uses it to route queries to a suitable model tier, which Google says cuts latency for everyday queries by up to 40%. Google also says Stitch uses it for rapid wireframes, Google Labs' CC uses it to choose emoji reactions, and Google Search uses it as a safety layer for MCP tools. Google says integrations are rolling out with OpenRouter, LangChain, Vercel, LiteLLM and others, with support for Gemini in Google Cloud coming soon.
How does it compare with similar products?
Google is not the first to offer this kind of model. The Decoder reports that after the startup Jev introduced a decision model, OpenAI, Cloudflare and Microsoft followed with similar products in recent weeks. Developer Simon Willison writes that Google's API is almost identical to Jev's, supporting the same three question types, and that its price matches OpenAI's GPT-6 Luna decision model. According to Willison, Jev is cheaper at 4.2 cents per million tokens but handles only text and JSON, while OpenAI's model accepts text and images and Google's covers text, images, audio, video and PDFs.
In a single informal test with one of his videos, Willison found that Google's model returned probabilities short of certainty rather than the flat 1.0 or 0.0 he says Jev and OpenAI return for yes/no questions.
Questions readers are asking
What is a decision model?
It is a model that scores a fixed set of possible answers, such as yes or no, instead of writing free text. Google says Gemini Decision 1 returns probabilities and confidence scores for each option in a single pass.
How much does the Gemini Decisions API cost?
Google says it costs $0.10 per million input tokens, output tokens are free, and a no-cost tier is available in Google AI Studio for prototyping.
Which inputs does it accept?
Text, images, audio, video and PDFs, with inputs of up to 256K tokens, according to Google.
Are the benchmark claims independently verified?
Not in the sources we reviewed. The benchmark and latency figures come from Google; NextFish has not independently verified them.
Sources
- #gemini
- #decision-models
- #api
- #developer-tools
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
AI2 min
HUMAIN sets Dec. 15 for its AI stack, promises a 3T-parameter model
The Saudi AI company's CEO calls the unnamed model the world's largest open-weight model and says it is far cheaper to run. No model card, licence or benchmark has been published.
Agents3 min
Google opens Antigravity's /teamwork agents, claims math and code results
Google Research says its multi-agent system helped prove a 2013 conjecture and port large codebases to Rust, and offers 10,000 academics free access. Results are Google's own.
Cybersecurity4 min
Cursor says key exposed at Fireworks led to malicious Claude Code extension
Cursor's incident report says attackers used stolen credentials to reach some SaaS tools and customer data, then planted a tampered extension up to 5,000 users may have received.
Cybersecurity3 min
Google Project Zero cuts bug disclosure deadline to 30 days by 2028
The Google team says AI has sped up attacks. It drops the extra 30-day patch soak time now and will cut vendors' deadline from 90 to 60, then 30 days.