🔍 Read the full analysis: Gemini Decisions API: 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, to evaluate up to 1,000 defined questions against one shared multimodal input. Google reports 150–250 milliseconds of end-to-end latency and input pricing of $0.10 per million tokens, but the source provides no independent benchmark results.
Google announced the Gemini Decisions API on Nov. 11, a service powered by its Gemini Decision 1 model that can score up to 1,000 yes/no, multiple-choice or rubric questions against one shared input. The API is designed for developers building systems that need to make several bounded decisions about the same material, rather than generate a separate text response for each question, as explained in the original analysis. Google reports 150–250 milliseconds of end-to-end latency; independent performance results are not included in the source material.
The endpoint accepts text, images, audio, video and PDFs, with Google saying a shared input can contain up to 256,000 tokens. It supports three question formats: yes/no conditions, selection among two to 255 options, and ordered rubric scores. Google says a request can include as many as 1,000 independent questions about the same input.
Responses include a confidence value from 0 to 1. Multiple-choice questions also return normalized probabilities for each option, while rubric questions return probabilities for each level and a weighted score. These values express the model’s assessment; on their own, they do not establish that a decision is correct or safe.
Google lists pricing at $0.10 per million input tokens, with no charge for output tokens. The company says the shared input is billed once per request, regardless of the number of questions, and each additional question adds only a small 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.
Batching Decisions Could Cut Repeated Calls
The API targets workflows where many checks apply to one shared item and the possible outcomes are already defined. A system might inspect a document alongside photographs, for example, and ask several policy or routing questions without sending the same material through separate calls. If the reported pricing and latency hold in customer workloads, batching could reduce repeated processing and make decision steps faster.
That could matter for request routing, policy screening and evaluation, where a system selects a category or action rather than composing an open-ended answer. Confidence scores could also help developers flag uncertain cases for a different model or a human reviewer. But the usefulness of those scores depends on how well they are calibrated for a particular task and dataset; a high confidence value is not a guarantee of a correct result.
Google says its Gemini Notebook product uses the endpoint to route queries and reports up to 40% lower latency for everyday queries. It also describes use in Stitch, CC from Google Labs and a safety layer for MCP tools in Search. These are company-reported examples, not independently verified customer results, and the source does not provide detailed measurements for each deployment.
Products Worth Considering
Samsung 32 Odyssey G5 (G51F) QHD 180Hz Gaming Monitor
Bose Lifestyle Ultra Soundbar and Ultra Subwoofer Bundle
Apple Watch Series 12 GPS 42mm Smartwatch
As an affiliate, we earn on qualifying purchases.
A Different Pattern for Model APIs
Many generative model APIs produce answers by generating text token by token. Google presents Gemini Decision 1 as a different approach: it scores candidate options in a single forward pass, making the endpoint a purpose-built alternative when an application has a finite set of possible outcomes. The distinction is about the task the API is designed to perform, not evidence that it will outperform other systems in every use case.
Google compares the Decisions API with Gemini 3.5 Flash-Lite used 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 provide enough information about workloads and cost assumptions to reproduce the comparison. The stated latency figures—150–250 milliseconds end to end, including 50–100 milliseconds attributed to model computation—are also Google’s reported figures, not independent findings.
The company says latency remains nearly flat from one to 100 questions per request and that its performance description includes multimodal inputs and contexts up to 16,000 tokens. That performance claim sits alongside the stated maximum shared input of 256,000 tokens; the source does not explain performance at the maximum context length.
“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 Is Limited
The source does not provide benchmark names, datasets, scores, confidence intervals or independent evaluations. It also does not explain how Google measured calibration or whether confidence values remain reliable across different tasks, modalities and customer data. Those gaps make it difficult to judge the model’s accuracy or compare the reported cost and speed with other approaches.
Details are also missing on performance across each supported modality, latency under heavier production traffic and how the service handles ambiguous inputs. Google describes requests with up to 1,000 questions, but the source gives no practical performance breakdown for larger batches. The nearly five-times-cheaper comparison lacks sufficient methodology to reproduce, while the reported Notebook latency improvement is not accompanied by a detailed measurement method.
These limits are particularly relevant if developers plan to use the API in safety checks or other consequential workflows. The announcement does not establish that model confidence predicts correctness in those settings. Teams would need to test error rates and calibration on their own cases before relying on the scores to automate decisions.
Testing Against Real Workloads
Developers can begin testing through the Gemini API and Google AI Studio, including the studio’s no-cost prototyping tier. A practical evaluation would compare response times and costs with existing classifiers or generative-model calls, then measure how often the API’s confidence scores identify cases that need review.
Further useful evidence would include published benchmark methods, calibration results and modality-specific performance data, as well as production experience from customers. Google has not announced a date for additional benchmark disclosures or a further product milestone in the source material. Until more detail is available, the speed, pricing and cost-comparison figures should be treated as company-reported launch claims.
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 scores and probabilities rather than requiring a separate generated text answer for every question.
How much does the API cost?
Google lists a price of $0.10 per million input tokens and no charge for output tokens. The company says shared input is billed once per request; the source does not provide independent confirmation of the pricing claims.
What latency does Google report?
Google reports 150–250 milliseconds end to end, with 50–100 milliseconds attributed to model computation. These are company-reported figures, and the source does not include an independent benchmark or full testing methodology.
Do the confidence scores guarantee an answer is correct?
No. A confidence score from 0 to 1 represents the model’s assessment, not proof of correctness. The source does not provide detailed calibration results showing how reliably those scores predict accuracy across tasks or customer data.
Where can developers try the API?
The endpoint is available through the Gemini API in Google AI Studio. Google AI Studio also has a no-cost tier for prototyping, according to the source material.
Primary source: Google AI · via ThorstenMeyerAI.com
