Google Launches Gemini Decisions API: The Fast, Low-Cost 'Judgment-Only' AI for Smarter Workflows

Google Launches Gemini Decisions API: The Fast, Low-Cost 'Judgment-Only' AI for Smarter Workflows

Codex4 min read78 views

On November 11, 2026, Google officially entered the emerging market of "judgment-only" AI models with the release of the Gemini Decisions API. Built upon the specialized Gemini Decision 1 model, this new offering is designed not to generate text, but to provide high-speed, accurate decisions for automated workflows.

This release follows the trend started by TypeSafe AI's "Jev" model, which gained attention for its ability to return classifications and scores rather than long-form content. Google's entry into this space brings multimodal capabilities and massive scale to the concept of decision-oriented AI.

Gemini Decisions API Overview

What is the Gemini Decisions API?

The Gemini Decisions API is a specialized tool provided by Google through Google AI Studio and soon Google Cloud. Unlike traditional Large Language Models (LLMs) that predict the next token in a sentence, Gemini Decision 1 evaluates a set of provided options in a single "forward pass." It returns a probability score (0 to 1) for each choice, making it ideal for tasks where the answer is constrained to a specific set of possibilities.

Key Use Cases

  • Tool Selection: Deciding which API or tool an agent should use next.
  • Policy Checking: Determining if a user's input violates safety or business guidelines.
  • Model Routing: Evaluating the complexity of a prompt to decide whether to send it to a small (cheap) or large (expensive) model.
  • Content Evaluation: Scoring the quality or relevance of information based on pre-defined criteria.

Technical Specifications and Question Types

Google has structured the API to handle three distinct types of decision-making tasks, which closely mirror the industry standards set by earlier models like Jev:

  1. noul: A binary Yes/No judgment.
  2. choice: Selecting one option from a list of 2 to 255 choices.
  3. score: Rating an input based on an ordered set of criteria.

One of the most impressive technical feats of Gemini Decision 1 is its efficiency. A single request can process up to 1,000 independent questions simultaneously against a shared input of up to 256,000 tokens.

Gemini Decisions Architecture

Performance and Speed

Latency is a critical factor for decision models. Google reports an end-to-end latency of 150 to 250ms, with the actual model calculation taking only 50 to 100ms. Remarkably, this latency remains nearly constant whether you are asking one question or one hundred, allowing for massive parallel processing of logic.

Multimodal Advantage: PDF, Video, and Audio

While competitors like Jev have focused primarily on text, Google’s Gemini Decisions API leverages the multimodal strengths of the Gemini family. It supports:

  • Text
  • Images
  • Audio
  • Video
  • PDF documents

This allows developers to build systems that can "watch" a video and decide if it contains specific content, or "read" a 200-page PDF to determine if it meets legal compliance standards, all within milliseconds.

Pricing: Disrupting the AI Market

Google has positioned Gemini Decision 1 as a cost-effective alternative to using full-scale generative models for simple classification tasks.

| Feature | Gemini Decisions API | TypeSafe AI Jev | | :--- | :--- | :--- | | Input Price (per 1M tokens) | $0.10 (approx. ¥16) | $0.042 (approx. ¥7) | | Output Price | Free | Free | | Modality Support | Full (Text, Image, Video, PDF) | Primarily Text (Experimental Image) | | Max Questions per Request | 1,000 | 100 | | Latency (Median) | 50-100ms (Model) | Ultra-low |

While Jev remains cheaper per token, Google’s multimodal support and the ability to batch 1,000 questions per request provide significant value for complex enterprise applications.

How to Get Started

As of November 11, the Gemini Decisions API is available through the Gemini API in Google AI Studio. Google is also working on integrations with popular developer tools, including:

  • OpenRouter
  • LangChain
  • Vercel

Google is already utilizing this technology internally for Google Search and Gemini Notebooks to improve response speed and routing efficiency. For developers looking to experiment, Google AI Studio offers a free tier for prototyping.

Frequently Asked Questions

Q: Is Gemini Decision 1 a replacement for Gemini 1.5 Pro? A: No. It is a complementary model. Use Gemini 1.5 Pro for generating content and Gemini Decision 1 for the logic, routing, and filtering that happens "behind the scenes."

Q: Does this model generate chat responses? A: No, it only returns scores and probabilities for predefined choices. It cannot "talk" to a user.

Q: Why is the output free? A: Because the model does not generate long sequences of text (tokens), the computational cost of the output is negligible compared to the input processing.

By focusing on "judgment" rather than "generation," Google is providing developers with a surgical tool to make AI agents faster, cheaper, and more reliable.