Gemini Decisions API is a new developer tool from Google that enables fast, probabilistic choices across text, images, audio, video, and PDFs. It evaluates up to 1,000 yes/no, multiple-choice, or rubric questions in one pass, providing a cheaper and faster alternative to generative text for routing and grading blog.google.
Most AI tools work by chatting—they write a response one word at a time. That process is often slow and expensive if you only need a simple 'yes' or 'no' or a choice from a list.
The Gemini Decisions API changes this by treating AI as a scoring machine rather than a writer. It can look at a video, a PDF, and a text note all at once and answer a thousand specific questions about them in a fraction of a second blog.google.
By the end of this guide, you will see how this cuts costs for app builders and why your favorite AI tools are about to feel much snappier.
Key takeaways
- Gemini Decision 1 scores options in a single pass instead of generating text token-by-token.
- It handles text, images, audio, video, and PDFs simultaneously up to 256K tokens.
- Costs are reduced to $0.10 per 1M input tokens with zero charges for output tokens.
- You can now use it via Google AI Studio or integrations like LangChain and Vercel.
What Gemini Decisions API actually does
Gemini Decisions API moves away from the traditional 'chatbot' style of AI. Instead of asking the AI to write a paragraph explaining its choice, this tool allows you to provide a set of candidate options and asks the AI to score them blog.google.
It supports three specific decision types: 'noul' for yes/no probabilities, 'choice' for selecting from up to 255 options, and 'score' for grading against a rubric. Because it doesn't have to 'write' the answer, it returns a confidence score from 0 to 1, allowing developers to send low-confidence results to a human reviewer for a final check.
This works across multiple formats at once. For example, you can feed the API a call recording and account metadata together, then ask it to determine if the customer is frustrated or if a specific policy was followed blog.google.
Gemini Decision 1 scores your candidate options in a single forward pass, returning probabilities and confidence scores
Google, blog.google
At a glance
| Detail | What we know |
|---|---|
| Model Name | Gemini Decision 1 (gemini-decision-1) |
| Latency | 150-250 ms end-to-end |
| Price | $0.10 per 1M input tokens; $0 output tokens |
| Capacity | Up to 1,000 independent questions per request |
| Input Limit | Up to 256K tokens |
| Supported Formats | Text, images, audio, video, and PDFs |
| Access Point | Google AI Studio |
Why Gemini Decision 1 is faster and cheaper
Gemini Decision 1 is built for speed because it skips the 'decoding' phase where AI models typically generate text token by token. This results in an end-to-end latency of 150-250 milliseconds, and the speed remains nearly flat whether you ask one question or 100 questions in a single request blog.google.
The pricing model is a significant shift for those using the Token and cost calculator. Google charges $0.10 per 1 million input tokens and bills zero for output tokens. According to Google, this makes the Decisions API nearly five times cheaper than using Gemini 3.5 Flash-Lite via the Interactions API for similar benchmark tasks blog.google.
By eliminating separate costs for OCR (optical character recognition) and audio transcription, the API simplifies the bill for multimodal projects.
AI generates text answers word-by-word, which is slow and bills for every output token.
AI scores pre-defined options in one pass, returning probabilities with zero output token costs.
Apps will respond instantly to complex multimodal inputs and become much cheaper to operate.
5x
5x cheaper across shared benchmarks than Gemini 3.5 Flash-Lite with the Interactions API.
- Exact performance limits for very large PDFs near the 256K token cap
- Availability of the no-cost tier for commercial production use
How Google uses the Decisions API internally
Google is already using the Decisions API as a 'control plane' to make other apps feel more responsive. In Gemini Notebook, the API acts as a real-time router, instantly deciding which model tier or 'thinking level' is best for a user's query. This has reduced latency for everyday queries by up to 40% blog.google.
Other applications include the Stitch rapid canvas experiment, where the API generates playable wireframes in seconds and applies voice or tap edits almost instantly. Even the emoji reactions in 'CC' from Google Labs use this API to pick the right emoji from a palette of 200+ options without adding lag blog.google.
Finally, Google Search employs the API as a safety layer, detecting adversarial behavior and content with high efficacy while staying within strict speed limits blog.google.
Hype vs reality
Instant AI decision making
Latency is 150-250ms, which is very fast but not literally instant.
How it compares
| Feature | Decisions API | Interactions API |
|---|---|---|
| Processing Style | Single forward pass | Token-by-token decoding |
| Output Cost | $0 | Billed per token |
| Latency | 150-250ms | Higher (variable) |
| Output Type | Probabilities/Scores | Generative Text |
What it means for you
Small business owners
You can automate policy checks or content grading across PDFs and images for pennies.
Do this: Explore the no-cost tier in Google AI Studio for prototyping.
Content creators
Tools you use for video or audio analysis will likely become faster and cheaper.
Do this: Check if your favorite AI tool has updated to Gemini 1.x models.
App developers
You can route users to different models instantly based on intent.
Do this: Review the Python and JavaScript guides in the Decisions API documentation.
Who should care (and who can skip it)
Worth your time if you…
- You build apps that need fast 'yes/no' answers
- You process large volumes of multimodal data
- You want to lower your AI API bill
You can skip it if you…
- You need the AI to write long-form creative text
- You only use AI via a chat interface
How to try the Decisions API
- Get an API Key Visit Google AI Studio and generate a unique API key.
- Choose your tool Select from Python, JavaScript, or REST guides in the official documentation.
- Test a Demo Try the 'Doorstep Parcels' or 'Speed Sketch' demo apps to see the scoring in action.
- Set your criteria Define your yes/no or multiple-choice options to avoid generative text output.
This simulates how you define a 'choice' request for the API.
Input: [UPLOAD PHOTO OF RECEIPT]. Question Type: choice. Options: [FOOD, TRANSPORT, OFFICE SUPPLIES, MISC]. Criteria: Select the category that best describes the majority of the spend on this receipt.
Use the 'confidence score' to set a threshold (e.g., 0.8). If the AI is less than 80% sure, have the system automatically flag the item for a human to review.
Trying to use this API for open-ended questions. It is designed for closed-set decisions; for long answers, stick to the standard Gemini models.
What to watch next
- Google Cloud rollout Support is coming soon to Gemini in Google Cloud, making it easier for enterprises to scale.
- Integration updates Watch for updates in LangChain and Vercel to simplify how you implement these decisions.
Our take
The Gemini Decisions API is a smart, surgical tool for developers who are tired of paying for 'chatty' AI when they just need a classification. It significantly lowers the barrier for building fast, multimodal apps. The biggest caveat is that it is strictly for closed-set decisions, so it cannot replace generative AI for content creation.
If you use an AI tool for grading or routing, check if they've switched to a 'decision' model to save on costs.
Your next step
- Free toolToken & Cost Calculator
Compare your current costs with the $0.10 per 1M input token rate.
- GuideAI models guide
Understand where Gemini Decision 1 fits compared to Flash or Pro.
- Free toolVideo to Prompt
Questions people ask
Is the Gemini Decisions API free?
The Gemini Decisions API has a no-cost tier for prototyping in Google AI Studio. For production, it costs $0.10 per 1 million input tokens, with no charges for output tokens.
What is the difference between this and a normal Gemini prompt?
Gemini Decisions API uses the Gemini Decision 1 model to score pre-defined options in one pass rather than generating text token-by-token. This makes it much faster and cheaper for specific choices.
Can it handle video and audio?
Yes, the Gemini Decisions API is natively multimodal and can process text, images, audio, video, and PDFs in a single request.
How many questions can I ask at once?
You can evaluate up to 1,000 independent yes/no, multiple-choice, or rubric questions against a single multimodal input in one pass.
Sources
- blog.google – Gemini Decisions API: Fast multimodal decisionsUsed for: All claims regarding pricing, latency, and technical capabilities.Official
Sources checked . We update this story when details change.
Written with AI assistance from the sources above and checked against them before publishing. Spotted an error? Tell us and we will fix it.