Introduction

On November 12, 2026, Google announced a fundamental pivot in its AI strategy with the release of Gemini Decision 1. While the industry has spent the last several years obsessed with the nuances of generative prose and creative reasoning, Google is now targeting the massive, untapped market of structured, high-speed decision-making. This isn't just another LLM; it is a specialized tool designed for the 'System 2' tasks of the enterprise world.

Gemini Decision 1 is launched alongside the new Decisions API, providing a dedicated pathway for developers who need answers, not essays. Instead of generating long-form text that requires further parsing, this model is purpose-built for closed-set decisions. It represents the third pillar of the Gemini family, standing alongside the existing frontier text models and high-fidelity image models, specifically optimized for classification, compliance, and structured evaluation.

Release Date: November 12, 2026
Core Function: Closed-set multimodal decision making
Primary Output: Calibrated probabilities and structured answers
Key Innovation: Zero token-generation overhead

Key Features & Architecture

The architecture of Gemini Decision 1 deviates significantly from standard autoregressive transformer models. Rather than predicting the next token in a sequence to form a sentence, the model is optimized to evaluate inputs against a predefined set of typed questions. This includes true/false, multiple-choice, and rubric-based queries. By bypassing the generative decoding stage, the model eliminates the latency and cost associated with output token generation.

Multimodality is baked into the core of the decision engine. The model can ingest and reason across text, images, video, audio, and complex PDF documents simultaneously. This allows for complex cross-modal reasoning, such as determining if a specific safety violation occurs in a video clip based on a written compliance rubric, or verifying if an audio transcript matches the visual cues in a document.

Input Modalities: Text, Image, Video, Audio, PDF
Question Types: True/False, Multiple Choice, Rubric-based
Decision Throughput: Up to 1,000 questions in a single pass
Output Format: Structured data with calibrated confidence levels

Performance & Benchmarks

In early benchmarks, Gemini Decision 1 has demonstrated a massive efficiency advantage over general-purpose models. According to the Artificial Analysis Decision Index v2.1, the model launched in the ninth position out of 18 specialized models, showing immediate competitiveness in reasoning and document review. Its ability to process up to 1,000 questions in a single API call sets a new standard for batch evaluation workloads.

Google Releases Gemini Decision 1 Multimodal Decision Model

Introduction

Key Features & Architecture

Performance & Benchmarks

API Pricing

Use Cases

Getting Started

Sources