Google Launches Gemini Decision 1: No Word-by-Word Generation, Evaluates 1000 Questions in One Request

Author: Dongmei, AI Frontier

While the entire AI industry is still competing over who can generate more tokens and who has stronger reasoning capabilities, Google has introduced a new model that does not rely on word-by-word generation to produce answers.

Google has just officially released Gemini Decision 1 and introduced the Gemini Decisions API, powered by the model. Unlike conventional generative large models, this model is designed for fast decision-making. It does not generate text token by token; instead, it scores predefined candidate answers through a single forward pass.

Google CEO Sundar Pichai also posted on X to introduce the key capabilities of Gemini Decision 1. He wrote:

Today, we are officially launching the Gemini Decisions API in Google AI Studio, powered by the new Gemini Decision 1 model.

Unlike conventional models that generate text token by token, this model is designed specifically for decision-making tasks over closed sets of options. In a single processing pass, it can score up to 1000 questions across text, images, audio, video, and PDFs.

Here are several details worth noting:

• Supports three decision types: yes/no judgments, multiple choice (up to 255 options), and rubric-based scoring;
• The price is only $0.10 per million input tokens, with zero fees for output tokens, making costs about one-fifth of Flash-Lite;
• It leads on multiple benchmark evaluations while having the lowest calibration error.

Google internal teams have already started using this technology.

He also demonstrated Stitch Rapid designing a complete product flow in seconds, with the video played at normal 1x speed.

This introduction reveals the biggest difference between Gemini Decision 1 and conventional large models: it is not responsible for stating the answer, but for making judgments from existing options.

Previously, when developers wanted AI to determine whether an email was spam or decide which Agent should handle a task, they often needed to call a generative large model and then extract the result from the generated text. Gemini Decision 1 eliminates the step of word-by-word decoding.

Developers only need to provide the input content, the question to be judged, and the candidate options; the model can directly return the corresponding judgment result, probability, and confidence.

This means that for tasks that only require a classification, score, or choice result, AI no longer needs to complete an entire text-generation pipeline.

According to data published by Google, Gemini Decision 1 can process up to 1000 independent questions in one request, supports multimodal inputs including text, images, audio, video, and PDFs, and has an end-to-end latency of only 150 to 250 milliseconds.

The pricing is also aggressive: only $0.10 per million input tokens, with no charge for output tokens. In the same set of benchmarks published by Google, its invocation cost is nearly 5 times lower than Gemini 3.5 Flash-Lite.

More notably, this is not just a model for answering multiple-choice questions. Google has already used it in scenarios such as AI model routing, Agent tool selection, product design, and security detection, aiming to make it a foundational component in AI systems dedicated to judgment and decision-making.

Processes 1000 Questions in One Pass, Supports Up to 255 Options

According to the technical introduction published by Google, Gemini Decision 1 offers three core decision-making capabilities.

The first type is Yes/No (binary judgment). For example, determining whether an image contains a safety risk or whether a piece of text violates established rules. The model provides the probability that the corresponding condition is true.

The second type is Multiple Choice. Developers can preset 2 to 255 candidate options; the model not only selects the answer that best matches the conditions but also returns normalized probabilities for each candidate.

The third category is Ordered Rubric (ordered criteria-based scoring). Developers can define scoring criteria for different levels, have the model evaluate input content, and return continuous weighted scores and probabilities for each level.

The three types of tasks can also be combined in a single request.

Google provides a simple example: upload a photo of an office pantry counter to the model and include a text prompt asking it to simultaneously determine whether there is a liquid spill on the counter, what cleanup measures should be taken, and the severity of the situation.

Traditionally, this may require designing multiple classifiers or executing different prompt tasks with generative large models. Gemini Decision 1 can complete these judgments in a single request.

In addition, the model supports multimodal input contexts of up to 256K tokens. For enterprise scenarios, this means developers can analyze call recordings together with customer profiles, or combine insurance claim documents with on-site photos to make judgments.

Each decision provides a confidence score from 0 to 1, making it easy for systems to route uncertain results to other models or human review.

Another competitive advantage of Gemini Decision 1 is speed and cost.

According to official Google information, in benchmark evaluations covering agent routing, reasoning, ranking, and multimodal understanding, Gemini Decision 1 delivers leading performance and the lowest calibration error.

In test contexts of no more than 16K tokens, the median latency in the model computation stage remains at 50 to 100 milliseconds, with end-to-end latency of approximately 150 to 250 milliseconds. Even when the number of questions in a single request increases from 1 to 100, latency remains largely stable.

On the common benchmark test set used officially, the cost of the Gemini Decisions API is nearly 5x lower than calling Gemini 3.5 Flash-Lite through the Interactions API.

7 first-place finishes across 9 tests, with accuracy up to 92.2%.

In this release of Gemini Decision 1, Google also published a noteworthy model comparison scorecard.

Unlike the usual practice of comparing general-purpose large models, Google selected three competing products for this release: OpenAI's GPT-6 Luna, TypeSafe Jev 1.14, which is specifically designed for classification decision-making, and its own Gemini 3.5 Flash-Lite.

The results show that Gemini Decision 1 ranks first in 7 of the 9 benchmark tests and also leads in probabilistic calibration. However, competitors performed better in the other two tests and in total invocation cost.

In the JevBench test, Gemini Decision 1 achieved 92.2% accuracy, surpassing GPT-6 Luna's 90.9%, TypeSafe Jev 1.14's 90.0%, and Gemini 3.5 Flash-Lite's 86.6%.

In the Amazon ESCI product search relevance test, Gemini Decision 1 achieved a 58.7% Macro-F1 score, 4 percentage points higher than GPT-6 Luna and 11.5 percentage points higher than Flash-Lite.

In the New Yorker Caption test, the gap widened further. Gemini Decision 1 scored 79.7%, while GPT-6 Luna scored 68.4%, a difference of 11.3 percentage points. This test involves understanding and selecting cartoon captions, assessing the model's contextual understanding and ability to judge candidate content.

In the PubMedQA medical question-answering test, Gemini Decision 1 also ranked first with a score of 79.8%, higher than GPT-6 Luna's 74.4%.

The results for the multilingual Global-MMLU-Lite test are similar. Gemini Decision 1 scored 86.0%, while GPT-6 Luna scored 79.6%, a difference of 6.4 percentage points.

However, Google has not achieved a lead across the board.

In the Mind2Web Multimodal test, TypeSafe Jev 1.14 ranked first with 86.8%, while Gemini Decision 1 scored 86.2%, trailing by 0.6 percentage points.

In the When2Call test, which evaluates tool-calling judgment, TypeSafe Jev 1.14 showed a clearer advantage, scoring 82.0%, compared with 79.9% for Gemini Decision 1.

This result is particularly noteworthy. Tool selection and timing decisions are among the key Agent scenarios targeted by Gemini Decision 1, yet in this test, the specialized decision model TypeSafe Jev still maintained the lead.

In addition, in the audio-understanding test MMAU and the video-understanding test Video-MME, Gemini Decision 1 scored 71.8% and 69.1%, respectively, both higher than Flash-Lite.

More noteworthy than accuracy is the probability calibration capability that Google specifically emphasized this time.

In the Top-1 ECE (Expected Calibration Error) test, Gemini Decision 1 had a calibration error of only 0.0250, while GPT-6 Luna was 0.0721.

In the same comparison, TypeSafe Jev 1.14 was 0.0494, while Flash-Lite reached 0.2104. In other words, Gemini Decision 1's calibration error was about 65% lower than GPT-6 Luna's and about 88% lower than Flash-Lite's.

This metric is especially important for Agent systems.

For example, when an Agent needs to determine whether a piece of code contains a security vulnerability, the model does not only provide a result of 'present' or 'absent'; it may also include a probability value. Based on the relationship between that probability and actual accuracy, the system can decide whether to take automatic action or escalate the case for human review.

Google is already using it: Agent routing latency reduced by up to 40%

Unlike some new models that only showcase benchmark scores, Google also disclosed use cases of the Gemini Decisions API in internal products.

The most noteworthy is Gemini Notebook.

Google is using the Decisions API as a real-time model router, selecting the appropriate model tier and reasoning effort based on the complexity of user requests. For routine simple questions, the system can prioritize lower-cost, faster-response models; for complex reasoning tasks, it routes requests to more capable models.

Google said that with this dynamic routing approach, latency for routine queries can be reduced by up to 40%.

This design is closely related to current trends in Agent systems. When completing a task, an Agent often needs to continuously determine the next action, such as which tool to select, which model to call, or whether human intervention is needed.

If every simple decision were handled by a large generative model, the system would incur additional latency and cost. Specialized decision models offer an alternative: route relatively simple, clearly defined judgments to low-latency models, while leaving complex tasks to general-purpose large models.

In addition to Gemini Notebook, Google disclosed three other application directions.

In Google Stitch, the newly introduced experimental Rapid Canvas feature uses the Decisions API to generate interactive product flow wireframes in seconds and quickly respond to modification instructions issued by users via voice or clicks. After users confirm the design direction, they can move the page into the full Stitch workflow for further refinement.

In Google Labs' CC product, the Decisions API is used to quickly select the appropriate emoji reaction from more than 200 options. Internally at Google Search, Google also uses this API to build a security detection layer for MCP tools, identifying potentially adversarial behaviors and content.

Reference link:Gemini Decisions API official blog

Source: TechFlow