🔍 Read the full analysis: A Closer Look At Fast, Calibrated Multimodal Decisions In Gemini Decisions API 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 score up to 1,000 defined questions against one shared text, image, audio, video or PDF input. Google reports 150–250 ms end-to-end latency and input pricing of $0.10 per million tokens, but the source provides no independent benchmark results or full methodology.
Google announced the Gemini Decisions API on Nov. 11, introducing a Gemini Decision 1-powered endpoint that can evaluate up to 1,000 yes/no, multiple-choice or rubric questions against one shared multimodal input, as detailed in the original analysis. The product is aimed at developers building workflows with a defined set of choices, and Google reports 150–250 milliseconds of end-to-end latency; those performance figures have not been independently validated in the source material.
The endpoint accepts text, images, audio, video and PDFs, with shared input contexts of up to 256,000 tokens, according to Google’s public beta announcement. Instead of asking a model to produce a separate free-form response for every check, developers can submit a batch of questions about the same material. The supported formats include yes/no conditions, selection among 2 to 255 options, and ordered rubric scores.
Responses include confidence values from 0 to 1. For multiple-choice questions, the API also returns normalized probabilities for each candidate. Rubric answers include probabilities by level and a weighted score. These outputs can help an application decide whether to proceed with a proposed action or route a case elsewhere, but a confidence value alone does 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. It says the shared input is billed once per request regardless of question count, while each additional question adds a small amount. The endpoint is accessible through the Gemini API in Google AI Studio, which offers a no-cost tier for prototyping.
Google reports 150–250 ms end-to-end latency, including 50–100 ms for model computation, and says latency stays nearly flat from one to 100 questions per request. Its description of these performance figures includes multimodal inputs and contexts up to 16,000 tokens. The source does not supply full benchmark results or an independent test of those claims.
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.
Batch Scoring Could Cut Repeat Calls
The API offers a different pattern for software that needs to make several bounded decisions about the same document, recording or other input. A workflow might check a claim document and damage photos against multiple conditions, or compare a call recording with account details for several routing criteria. If the reported speed and pricing hold in production, one shared-input request could reduce the need for repeated model calls.
Google says its Gemini Notebook product uses the endpoint to route queries and reports up to 40% lower latency for everyday queries. It also cites use in Stitch, CC from Google Labs, and a safety layer for MCP tools in Search. These are company-reported deployments; the announcement does not provide independent customer measurements or detailed results for each use.
For teams considering the API in policy, safety or other consequential workflows, the key question is not only how quickly it returns scores. They will also need to test whether its confidence values are calibrated on their own data, how it handles ambiguous cases, and when human review or a separate fallback system is appropriate.
Products Worth Considering
Samsung 32-Inch Odyssey G5 QHD 180Hz Gaming Monitor
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
Amazon Kindle Scribe Colorsoft 64GB with 11-Inch Color Display and Premium Pen, Graphite
As an affiliate, we earn on qualifying purchases.
A Scoring Endpoint, Not a Chat Reply
Many generative model APIs return a response by producing text token by token. Google describes Gemini Decision 1 as scoring candidate options in a single forward pass, presenting the endpoint as a purpose-built tool for cases where the available actions or categories are already defined. The distinction is relevant for developers who need classifications or routing decisions rather than explanatory prose.
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 identify the benchmarks, publish scores, or give enough detail about the workloads and cost assumptions to reproduce that comparison. The figure should be treated as a company claim, not an independently established result.
The announcement also describes low calibration error, but the source material does not include the underlying measurements. It gives no benchmark names, datasets, scores, confidence intervals or outside evaluation. That limits what developers can conclude about performance beyond the use cases and figures Google reports.
“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 and Calibration Need Testing
The source does not provide independent benchmark results or the methodology behind Google’s latency, cost-comparison or calibration claims. It also does not show how results vary across individual supported modalities, beyond general performance descriptions. That makes it difficult to compare the product with existing classifiers or model calls using reproducible evidence.
Google says confidence scores range from 0 to 1, but the announcement does not explain how calibration was measured or whether the scores remain reliable across tasks and customer data. Details are also missing on performance under heavier production traffic, the practical handling of ambiguous inputs, and how latency changes when a request approaches the 1,000-question limit. These gaps matter most when teams may use the outputs in safety checks or other high-impact decisions.
The source gives no date for additional benchmark disclosures or for further product milestones. Until more evidence is available, the speed, calibration and comparative cost figures remain Google’s reported claims.
Developers Can Test Their Own Workloads
Developers can begin prototyping through the Gemini API in Google AI Studio, including its no-cost tier. The most useful next evidence would include published benchmark methods, calibration results, performance across modalities and measurements from production use.
Teams evaluating the endpoint can compare its response times and costs with their current classifiers or generative-model calls, using representative inputs and decision tasks. They can also check whether confidence scores reliably flag cases that need a fallback model or human review. Google has not announced a timeline for further benchmark disclosures in the source material.
Key Questions
What does the Gemini Decisions API do?
It evaluates batches of up to 1,000 defined questions against one shared input. Questions can be yes/no checks, choices among 2 to 255 options, or ordered rubric scores.
Which kinds of input does it support?
Google says the endpoint accepts text, images, audio, video and PDFs. It reports support for shared inputs of up to 256,000 tokens.
How much does the API cost?
Google lists a price of $0.10 per million input tokens and no charge for output tokens. It says the shared input is billed once per request, while additional questions add a small amount.
Are the speed and accuracy claims independently verified?
The source material provides no independent evaluation or full benchmark methodology. Google reports latency and calibration claims, but developers should test these against their own workloads.
What do the confidence scores mean?
The API returns confidence values from 0 to 1; choice questions also include probabilities for each option. The announcement does not explain how calibration was measured, and a score alone does not prove a decision is correct.
Primary source: Google AI · via ThorstenMeyerAI.com
