🔍 Read the full analysis: What Makes The Gemini Decisions API Fast And Calibrated For 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 scoring up to 1,000 defined questions against one shared multimodal input. Google reports 150–250 ms end-to-end latency and input pricing of $0.10 per million tokens, but the announcement does not provide independent benchmark results or detailed calibration methods.
Google announced the Gemini Decisions API on Nov. 11, as detailed in the original analysis, a service powered by its Gemini Decision 1 model that evaluates up to 1,000 yes/no, multiple-choice or rubric questions against one shared text, image, audio, video or PDF input. Google says the endpoint returns results in 150–250 milliseconds and charges $0.10 per million input tokens, but the announcement does not include independent performance or calibration tests.
The API is designed for tasks where a system must select among defined outcomes rather than generate an open-ended response, a focus shared by other fast, cost-effective Gemini models. Google says one request can contain input of up to 256,000 tokens and as many as 1,000 independent questions. Multiple-choice questions can offer between two and 255 options; rubric questions use ordered levels.
Results include a confidence value from 0 to 1. For choice questions, the API also returns normalized probabilities for each candidate. Rubric results include probabilities by level and a weighted score. Google describes the model as scoring candidate options in a single forward pass instead of generating an answer token by token.
Google lists pricing at $0.10 per million input tokens, with no charge for output tokens. It says the shared input is billed once per request, regardless of the number of questions, and that each question adds only a small amount to the cost. The endpoint 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.
The API’s proposed advantage is that a workflow can ask several bounded questions about the same material without making a separate generative-model call for each one. For example, a system might assess a document and accompanying photos against several policy conditions, or use a call recording and account information to route a request. If the reported performance and pricing hold in practice, that approach could reduce repeated processing in workflows with many related checks.
Confidence values could also help developers direct uncertain cases to another model or a human reviewer. But a confidence score is not evidence that a result is correct, safe or well calibrated for a particular organization’s data. Teams using the API in consequential decisions would need to test its outputs against their own cases and define what happens when scores are uncertain or inputs are ambiguous.
Google reports that Gemini Notebook uses the endpoint for query routing and has seen up to 40% lower latency for everyday queries. The company also cites use in Stitch, CC from Google Labs and a safety layer for MCP tools in Search. These are company-reported examples; the announcement does not provide outside customer evaluations or detailed measurements for each deployment.
Products Worth Considering
Amazon Kindle Paperwhite (2024, 16 GB) – 7-inch Display, Black
Samsung Odyssey OLED G5 27-Inch QHD Gaming Monitor
LG UltraGear evo 27-inch 5K Mini-LED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
A Purpose-Built Alternative to Generation
Many generative model APIs produce answers sequentially as text tokens. Google presents Gemini Decision 1 as a different pattern: a model for assigning scores to a set of candidate choices in one forward pass. That design is aimed at systems with a finite set of actions, categories or rubric outcomes, such as selecting a model tier, checking a policy condition or routing a request.
Google says the API’s latency is 150–250 milliseconds end to end, of which 50–100 milliseconds is attributed to model computation. It says latency stays nearly flat from one to 100 questions per request and describes performance with multimodal inputs and contexts up to 16,000 tokens. The same announcement gives a maximum context size of 256,000 tokens; it does not explain how the larger limit relates to the performance figures.
Google also compares the Decisions API with Gemini 3.5 Flash-Lite through its Interactions API, saying the new setup was nearly five times cheaper across a shared benchmark suite. The source does not name the benchmarks, publish scores or explain the workload and cost assumptions, so the comparison cannot be independently reproduced from the announcement.
““Rather than generating answer text token by token, Gemini Decision 1 scores your candidate options in a single forward pass.””
— Google, in its announcement
Performance Claims Await Independent Tests
The launch material does not provide independent benchmark results, named datasets, scores, confidence intervals or a detailed method for evaluating calibration. Google gives confidence values on a 0-to-1 scale but does not explain how calibration was measured or whether the scores remain reliable across different tasks, modalities and customer data.
The announcement also does not show how latency varies across each supported input type, how the service performs under heavier production traffic, or how the maximum of 1,000 questions works in practice beyond the stated limit. The cost comparison with Gemini 3.5 Flash-Lite lacks enough published detail to reproduce. These gaps matter most for teams considering the endpoint for safety checks or other high-impact decisions.
Production Testing Will Settle the Claims
Developers can begin testing through the Gemini API and the no-cost prototyping tier in Google AI Studio. A practical evaluation would compare the service with existing classifiers or generative-model calls on the same inputs, measure response times and costs, and check whether confidence scores identify cases that need review.
Further evidence would include published benchmark methods, calibration results by task and modality, and production experience from customers. The source material gives no date for additional benchmark disclosures or other product milestones, so those developments remain unannounced.
Key Questions
What does the Gemini Decisions API do?
It evaluates up to 1,000 defined questions against one shared input, which can include text, images, audio, video or PDFs. It returns choices or rubric results with probabilities and confidence values rather than relying on a separate generated text answer for each question.
How fast and how much does Google say it costs?
Google reports 150–250 milliseconds of end-to-end latency, with 50–100 milliseconds attributed to model computation. Pricing is listed as $0.10 per million input tokens, with no output-token charge; these are Google’s launch figures, not independent measurements.
Do confidence scores prove that an answer is correct?
No. A confidence value is an output score, not proof of correctness or safety. The announcement does not publish calibration methods or independent results, so developers would need to test whether scores are reliable for their own tasks and data.
Can developers try the API now?
Google says the endpoint is available through the Gemini API in Google AI Studio, which has a no-cost tier for prototyping. The announcement does not specify a date for further benchmark publications or product milestones.
Primary source: Google AI · via ThorstenMeyerAI.com
