Google Introduces Gemini Decisions API with Gemini Decision 1 at $0.10 per Million Tokens

12 November 2026 3 min read هدهد تك | Hudhud Tech

Google has launched the Gemini Decisions API, powered by a new model, Gemini Decision 1 (gemini-decision-1). The API does not generate text but scores up to 1,000 typed questions against a shared input, returning calibrated probabilities. This service is designed for developers to manage small branching calls within agents, such as routing, policy checks, tool selection, and grading. The pricing is set at $0.10 per 1 million input tokens, with no charges for output tokens.

TL;DR

Size: Context is 262,144 tokens (256K) across shared input and all questions.
Runs on: Google-hosted API only (Gemini Developer API, v1beta). No open weights or self-hosting.
Performance: Leads 7 of 9 accuracy benchmarks in Google’s table, with the lowest calibration error.
Best: 92.2% on JevBench (public split).
Worst: 58.7% Macro-F1 on Amazon ESCI, its lowest score. It also trails Jev 1.14 on When2Call (79.9% vs 82.0%).
Bottom line: Best: multimodal, calibrated decisions in 150-250 ms. Worst: not the cheapest; Jev 1.14 costs $0.68 vs $0.88 on the shared suite.

What is the Gemini Decisions API?

Gemini Decision 1 is a specialized multimodal model from Google designed for closed-set decisions. It processes text, images, video, audio, and PDFs in a single prefill pass. The output token usage is always zero, allowing code to branch directly based on the returned probabilities.

Each request consists of three parts: a model ID, a shared input, and a map of questions. The documentation defines three question types:

noul (yes/no): returns the probability of a yes answer, bounded between 0.01 and 0.99.
choice: selects one option from 2 to 255 and returns normalized probabilities for each option.
score: evaluates input on 2 to 255 ordered rubric levels, returning a probability-weighted score.

Each answer includes a confidence value ranging from 0 to 1. Google recommends routing low-confidence cases to a fallback or a human reviewer. Questions are scored in parallel and do not influence each other.

How does Gemini Decision 1 work?

Unlike standard models that decode answers token by token, Gemini Decision 1 scores candidate options in a single forward pass. Google reports median model compute times of 50-100 ms, with end-to-end processing taking 150-250 ms for inputs up to 16K tokens, remaining consistent from 1 to 100 questions.

Billing is structured similarly. The shared input is billed once per request, regardless of the number of questions asked. Each question incurs charges only for its own instruction and criteria tokens, with no additional fees for thinking tokens, transcription, or OCR.