🔍 Read the full analysis: How The Gemini Decisions API Delivers Fast, Calibrated Multimodal Decisions on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google announced the Gemini Decisions API on Nov. 11, powered by Gemini Decision 1, for evaluating up to 1,000 defined questions against one shared text, image, audio, video or PDF input. Google reports 150–250 milliseconds of end-to-end latency and input pricing of $0.10 per million tokens, but the announcement does not provide independent benchmarks or detailed calibration results.
Google announced the Gemini Decisions API on Nov. 11, as described in the original analysis, offering developers a way to evaluate up to 1,000 defined questions against a shared text, image, audio, video or PDF input in one request. Powered by Gemini Decision 1, the endpoint returns probabilities and confidence scores rather than generating a separate written answer for each question, a design Google says can support faster routing and classification workflows.
The API supports three question formats: yes-or-no checks, choosing among two to 255 options, and ordered rubric scores. A request can include shared input of up to 256,000 tokens, according to Google. For choice questions, the response includes normalized probabilities for each option; rubric results include probabilities by level and a weighted score. All answers include confidence values from zero to one.
Google reports 150–250 milliseconds of end-to-end latency, with 50–100 milliseconds attributed to model computation. The company says latency stays nearly flat between one and 100 questions in a request, and its performance description includes multimodal input and contexts up to 16,000 tokens. These are company-reported figures: the source material does not include independent tests or full benchmark results.
Google lists the price as $0.10 per million input tokens and says there is no charge for output tokens. It says the shared input is billed once per request, regardless of the number of questions, with each added question contributing only a small amount to cost. The API is available through the Gemini API in Google AI Studio, which offers a no-cost tier for prototyping.
Gemini Decisions API Fast, calibrated multimodal decisions
Google’s API scores up to 1,000 defined questions against shared multimodal input. Its speed, pricing and calibration claims await independent evaluation.
How it works
Gemini Decision 1 scores defined answers in one pass, instead of generating a separate text response for each question.
Provide text, images, audio, video or PDFs once for the batch.
Ask yes/no, multiple-choice or ordered rubric questions.
Get answers with confidence values; choice questions also return probabilities.
Where it could help
Several checks on the same material may need fewer repeated model calls.
Choose a path
Route requests to a model tier or fallback when confidence is low.
Check policy
Score defined conditions across documents, images or recordings.
Grade against a rubric
Return ordered scores for context without generating a separate answer for each check.
What the announcement supports
These figures and deployments are reported by Google; the source gives no independent evaluation.
- Latency stays nearly flat from one to 100 questions.
- Gemini Notebook reports up to 40% lower latency for everyday queries.
- Shared input is billed once per request.
- Benchmark names, datasets and detailed results.
- How confidence calibration holds across tasks and customer data.
- Performance across modalities and production workloads.
Try it in context
Developers can prototype through the Gemini API in Google AI Studio’s no-cost tier.
Batching Could Cut Repeated Model Calls
The API is aimed at workflows where a system must choose among known actions or categories, rather than produce open-ended text. A single document, recording or image set could be checked against several conditions at once—for example, a policy check, a routing decision and a rubric score. If the reported latency and pricing hold in production, evaluating those questions together could reduce repeated calls when many decisions rely on the same material.
Confidence values may help developers flag uncertain outputs for a fallback model or human review. But a confidence score is not proof that a result is correct or safe. Its value depends on how well it reflects actual error rates for a particular task and dataset. That distinction matters especially when teams use scores to affect customer access, safety reviews or other consequential decisions.
Google says Gemini Notebook uses the endpoint to route queries and reports up to 40% lower latency for everyday queries. The company also describes uses in Stitch, CC from Google Labs and a safety layer for MCP tools in Search. The announcement provides no outside customer results or detailed measurements for those deployments, so the figures and examples should be read as Google’s claims.
Products Worth Considering
Samsung Galaxy Tab A9+ 11-Inch Android Tablet, 64GB, Graphite
Samsung Galaxy Tab A11+ 8GB RAM, 256GB Storage, Gray
Amazon Kindle Scribe (3rd Generation), 64GB, 11-inch Display, Graphite, with Premium Pen
As an affiliate, we earn on qualifying purchases.
A Scoring API, Not Text Generation
Many generative model APIs produce responses as a sequence of tokens. Google describes Gemini Decision 1 as scoring candidate options in a single forward pass, positioning the endpoint for cases in which the possible answers are already bounded. Developers can send one shared input and request multiple evaluations, rather than asking a model to generate separate prose responses for each check.
Google also compares the Decisions API with Gemini 3.5 Flash-Lite through its Interactions API, saying the new approach was nearly five times cheaper across a shared benchmark suite. The source does not name the benchmarks, publish scores or explain the workloads and cost assumptions in enough detail to reproduce that comparison. The announcement likewise does not provide independent evidence for its performance claims.
“Rather than generating answer text token by token, Gemini Decision 1 scores your candidate options in a single forward pass.”
— Google, in its announcement
Benchmark and Calibration Evidence Pending
The announcement does not publish benchmark names, datasets, scores or confidence intervals, nor does it include an independent assessment of the claimed calibration performance. Google gives confidence values from zero to one but does not explain how calibration was measured or whether those values remain reliable across different tasks, modalities and customer data.
Several operating details are also missing. The source does not show how latency varies for each supported modality, how the endpoint performs under heavier production traffic, or how ambiguous inputs are handled. Although requests can include up to 1,000 questions, the reported near-flat latency comparison covers one to 100 questions; performance beyond that range is not specified. The cost comparison with Flash-Lite also lacks enough methodological detail for an outside reader to verify it.
Production Tests Will Show Its Value
Developers can start prototyping through the no-cost Google AI Studio tier and access the endpoint through the Gemini API. For teams considering deployment, the practical next step is to test their own inputs and question sets against existing classifiers or model calls, measuring response time and cost while checking whether confidence scores identify cases that need review.
Further documentation or evaluations could clarify benchmark methods, calibration across modalities and performance at production scale. Google has not announced a date for additional benchmark disclosures or another product milestone in the source material. Until those details emerge, the reported figures remain launch claims rather than independently verified results.
Key Questions
What does the Gemini Decisions API do?
It evaluates up to 1,000 defined questions against one shared input and returns results such as yes-or-no answers, option probabilities, rubric scores and confidence values.
What kinds of input does it support?
Google says the API accepts text, images, audio, video and PDFs. It describes shared inputs of up to 256,000 tokens, while its latency discussion refers to contexts up to 16,000 tokens.
How much does the API cost?
Google lists pricing at $0.10 per million input tokens and says output tokens are not charged. The company says shared input is billed once per request; developers should check current product pricing for applicable terms.
Are the speed and calibration claims independently verified?
Not in the source material. The 150–250 millisecond latency figure and other performance statements are attributed to Google, and the announcement does not provide independent benchmark results or detailed calibration methods.
Primary source: Google AI · via ThorstenMeyerAI.com
