Google Launches Gemini Decision 1 for 1,000 Instant Choices

Key Takeaways

1Google introduces Gemini Decision 1, a model that handles up to 1,000 questions in a single query

2Responses are probabilized with a confidence score between 0 and 1

3Announced latency of 150 to 250 ms and a price of $0.10 per million input tokens

4Human supervision required for sensitive cases and no automated sequential actions

Why it matters : This model aims to accelerate repetitive decision-making in software while imposing safeguards on reliability and orchestration.

Google offers a specialized model that evaluates up to 1,000 questions in a single query, providing probabilistic answers with a reported latency of around a hundred milliseconds. The tool targets repetitive decisions, but Google emphasizes the scope: no sequential actions, a need for orchestration on the software side, and human oversight when the stakes are high.

Required Safeguards and Extreme Performance Not Guaranteed

A probability estimate is not sufficient to resolve an ambiguous case: a fast AI can also make mistakes quickly. Developers must monitor errors and set appropriate confidence thresholds, with human validation whenever decisions could have significant consequences. The stability of processing time is not guaranteed when a thousand questions are accompanied by very large files.

Internal Deployments: Routing in Notebook and Security Control

Google is already using Gemini Decision 1 to distribute queries among multiple models within Gemini Notebook, which allows for a reduction in latency of up to 40% for the most frequent requests. This model is also utilized in the rapid prototyping of Google Stitch and for certain security checks associated with Google Search's MCP tools.

What the Model Does: Choosing from Options, with a Score from 0 to 1

The model provides probabilities on predefined answers rather than developed texts. Three formats are available: yes/no, multiple choice with 2 to 255 options, and scoring according to a grid, with each output accompanied by a confidence score between 0 and 1 to direct uncertain cases to a human. It can assist multi-tool AI agents but is limited to choosing from the provided options and does not, on its own, replace a business process.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Multimodal Inputs and Mass Evaluation of Up to 1,000 Questions

Gemini Decision 1 supports text, images, audio, video, and PDF files, with a context capacity of about 256,000 tokens. It can analyze up to 1,000 distinct questions in a single pass, without performing a thousand consecutive operations: when decisions are conditioned by previous steps, it is up to the software to manage these dependency relationships. Among the mentioned use cases is that of a customer service tasked with inspecting parts, assessing damages, or guiding a case.

Developer Access, Announced Latency, and Pricing at $0.10/M Tokens

The model is primarily aimed at developers via the Gemini Decisions API in Google AI Studio, with possible integration in Python, JavaScript, or via REST. Google reports a total latency of 150 to 250 ms, with 50 to 100 ms for model computation on inputs of up to 16,000 tokens, maintaining a nearly stable time between one and one hundred questions. The pricing is set at $0.10 per million input tokens, with no charge for output; shared data is counted once per request, and the instructions for each question add to the total. Tests published by Google suggest a cost nearly five times lower than Gemini 3.5 Flash-Lite on their evaluation set, a comparison that remains to be confirmed in real-world conditions. Introduced on November 11, 2026, the tool emphasizes rapid decision-making via an API for repetitive tasks.