🔍 Read the full analysis: Gemini Decisions API For Fast, Calibrated Decisions Across Modalities 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; independent benchmarks and detailed calibration results are not included in the announcement.
Google announced the Gemini Decisions API on Nov. 11, introducing an endpoint powered by Gemini Decision 1 that can score up to 1,000 defined questions against the same text, image, audio, video or PDF input. The launch gives developers a way to evaluate several bounded choices in one request rather than generate a separate text response for each, though the performance and cost figures published so far are Google’s claims and have not been independently assessed in the supplied source.
The endpoint accepts a shared input of up to 256,000 tokens and supports three question formats: yes/no conditions, selection from 2 to 255 options, and ordered rubric scores. Google says a request can contain up to 1,000 independent questions. The questions are evaluated against the same material, a structure intended for workflows that need multiple checks or classifications on one item.
Responses include confidence values from 0 to 1. For choice questions, the API also returns normalized probabilities for each candidate. Rubric questions provide probabilities by level and a weighted score. These outputs express the model’s scoring, but a confidence value alone does not establish that a result is correct or safe for a particular use.
Google reports 150–250 milliseconds of end-to-end latency, including 50–100 milliseconds 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 announcement does not supply independent tests or full benchmark results to verify those statements.
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, while each question adds a small amount to the cost. The endpoint is available through the Gemini API in Google AI Studio, which Google says 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.
Batch Scoring for Defined Choices
The API is designed for software that must choose among predefined actions or categories, such as routing a request, checking a policy condition, selecting a model tier or grading material against a rubric. When a workflow needs many judgments about the same document or recording, batching could reduce the need to send repeated model calls, if the service performs as Google describes.
Its multimodal input support could let developers run related checks on combined material—for example, reviewing a claim document alongside damage photos, or comparing a call recording with account information. Google also says Gemini Notebook uses the endpoint to route queries and reports up to 40% lower latency for everyday queries. The company describes use 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.
For teams considering the endpoint in consequential workflows, the confidence scores need practical validation. Developers would need to test whether the scores are calibrated on their own data and whether uncertain or high-impact cases should be sent to a human reviewer or another system. Speed and cost could matter, but they do not answer those reliability questions.
Products Worth Considering
Anker SOLIX C300 DC Portable Power Station with 60W Solar Panel, 288Wh, 300W
DJI Mini 4K Drone with 4K UHD Camera
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
A Scoring Alternative to Text Generation
Many generative model APIs produce answers as a sequence of tokens. Google describes Gemini Decision 1 as scoring candidate options in a single forward pass, rather than generating answer text token by token. That makes the endpoint a purpose-built alternative when an application already has a finite set of answers and needs scores for several questions about shared material.
Google 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 material does not name the benchmarks, publish scores or explain the workload and cost assumptions in enough detail to reproduce the comparison. The figure should therefore be treated as a company-reported comparison, not an independently verified result.
The announcement also describes the API as having low calibration error, but provides no benchmark names, datasets, calibration measurements or independent evaluation in the supplied material. That leaves a distinction between the feature’s stated output—probabilities and confidence values—and evidence that those values reliably track accuracy across tasks.
““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 Gaps
The launch information supplied does not include independent benchmark results, named datasets, published scores or confidence intervals. It also does not explain how Google measured calibration, or whether confidence scores remain reliable across different tasks, modalities and customer data.
Performance details are limited as well. The announcement gives general latency figures and a statement about contexts up to 16,000 tokens in its performance description, but does not show results broken down by modality or explain how latency changes under heavier production traffic. It is also unclear how the API behaves on ambiguous inputs or how practical limits apply when a request contains large numbers of questions.
The reported cost comparison with Gemini 3.5 Flash-Lite lacks enough information to reproduce it. Google has not, in the supplied source, provided outside customer evaluations or a date for further benchmark disclosures. Teams should treat the launch figures as vendor-reported claims until more methods and evaluations are available.
Testing Against Production Workloads
Developers can begin prototyping through the no-cost tier in Google AI Studio and access the endpoint through the Gemini API. For a useful evaluation, teams can compare its latency and costs with their current classifiers or generative-model calls using the same inputs and decision tasks.
They can also test whether confidence scores identify cases that need review, including on the data and modalities they expect to use in production. Further documentation or independent evaluations covering benchmark methods, calibration and modality-specific performance would help clarify how well the launch claims generalize. Google has not announced a schedule for those disclosures or other product milestones in the supplied source.
Key Questions
What does the Gemini Decisions API do?
It scores up to 1,000 yes/no, multiple-choice or rubric questions against one shared input, which can include text, images, audio, video or PDFs.
What are the announced price and latency?
Google lists $0.10 per million input tokens and no charge for output tokens. It reports 150–250 milliseconds of end-to-end latency, but the supplied source does not provide independent measurements.
Do the confidence scores prove that an answer is correct?
No. The API returns confidence values and probabilities, but those values do not by themselves establish correctness or safety. The announcement does not include detailed calibration results, so developers would need to test them on their own workloads.
Are the performance and cost comparisons independently verified?
Not in the supplied announcement material. Google reports the latency, pricing and nearly five-times-lower cost comparison, but independent benchmarks and enough methodological detail to reproduce the comparison are not provided.
Primary source: Google AI · via ThorstenMeyerAI.com
