Skip to content

Google joins Jev and OpenAI with Gemini Decisions API

Google's Gemini Decision 1 answers yes/no, choice and rubric questions from text, images, audio, video and PDFs, priced level with OpenAI's decision model and above Jev.

Draftship Editorial3 min read

The Gemini Decisions API arrived on 11 November 2026, with Google putting a model called Gemini Decision 1 into a small category of APIs built to answer questions rather than write prose. Simon Willison covered it in a link post on his weblog, placing Google alongside Jev and OpenAI, whose decision model he names as GPT-6 Luna.

Three question types and a borrowed word

Gemini Decision 1 supports the same three question types as Jev, according to the post: yes/no, choice, and rubric scoring. It also reuses Jev's term "Noul" for a yes/no question. The post does not explain what the word means or where it comes from, so treat the naming as inherited rather than descriptive.

Price and input types

Gemini Decision 1 is priced at 10 cents per million input tokens, with output free. That matches the GPT-6 Luna decision model exactly. Jev remains the cheapest of the three at 4.2 cents per million input tokens.

The wider difference is what each model accepts. Jev is text and JSON only. OpenAI's decision model handles text and images. Gemini Decision 1 is described as fully multi-model, covering text, images, audio, video and PDFs, which makes it the broadest of the three on input types.

The video test, and the confidence gap

Willison tested the model with a video file, described as his Yonkers stomping robot video, sent along with four yes/no questions: whether the video contains a dragon, whether it contains a robot, whether it is set in France, and whether it is set in the USA.

The answers came back as probabilities rather than hard labels. The robot and USA questions scored high, while the dragon and France questions scored low. Willison notes that Jev and OpenAI both return 1.0 or 0.0 for yes/no questions, and that Gemini insists on being less confident than that. The post shows the response JSON for yes/no answers only.

What the probabilities likely mean for your code

This section is inference, not something the source states. A probability instead of a binary answer suggests you can set your own cutoff rather than accept the provider's, so a classification check could act above a threshold you choose and log everything else. It also suggests you can see how close a call was, which a 0.0 or 1.0 hides. Whether those numbers are well calibrated is not something the post addresses.

The practical uses are the unglamorous ones: deciding whether a user's upload fits a category, routing a support message to the right queue, scoring a submission against a written rubric. The post only compares pricing inside the decision model category, so it makes no comparison against general purpose models doing the same job.

Limits and open questions

This is a link blog post, not documentation. It gives no launch status beyond the date, no availability regions, no rate limits and no latency figures. It shows the response shape for yes/no questions only, so choice and rubric-scoring output is not demonstrated. It does not say whether free output applies to every input type, or what the size limits are for video and PDFs. Anyone planning to depend on the API should read Google's own documentation rather than treat this post as a specification.

What it means for building apps with AI

Decision models are a reminder that not every AI call in an app has to generate text. Much of app logic is classification, and an endpoint that returns a label fits that work better than a chat completion. If you are wiring a provider like this into app behaviour, Draftship's guide to adding AI to your app covers how AI calls fit into app logic. Draftship is not affiliated with Google, Jev or OpenAI, and nothing here suggests it supports Gemini Decision 1 specifically.

Primary sources

This article was drafted with AI assistance from the primary sources listed below, then reviewed and approved by a Draftship editor.

AI APIs

Gemini Decision 1 lands on Vercel AI Gateway

Vercel added Google's Gemini Decision 1 to AI Gateway. The model returns booleans, choices and scores for routing, policy checks and ranking, without generating output tokens.