The Extended Brief

Gemini Decisions API

Brief by The AI News AI newsroom · Nov 12, 2026, 1:02 AM EST edition

Original reporting by Simon Willison · published Nov 11, 2026, 5:03 PM EST

Teams using decision-model APIs gain a third supplier — the only one that ingests video, audio, and PDFs — priced at 10 cents per million input tokens.

Key points

• Google's Gemini Decision 1 is the only one of the three decision-model APIs accepting audio, video, and PDFs. source ↗
• It matches OpenAI's GPT-6 Luna at 10 cents per million input tokens with free output; Jev costs 4.2 cents. source ↗
• The API supports the same three question types as Jev — yes/no, choice, and rubric-scoring — and reuses Jev's "Noul" term. source ↗
• Where Jev and OpenAI return exactly 1.0 or 0.0, Gemini returns fractional probabilities for yes/no questions. source ↗
• In the author's video test, Gemini scored "robot" at 0.92 probability and "dragon" at 0.05. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Rerun a representative slice of your decision-question workload on Gemini Decision 1 to compare answer quality against Jev or GPT-6 Luna before committing to a provider.
• If your pipeline needs video, audio, or PDF inputs, prototype on Gemini Decision 1, since neither Jev nor OpenAI's decision model accepts them.
• Verify your downstream code tolerates fractional probabilities, because Gemini returns values like 0.9234 instead of the hard 1.0 or 0.0 that Jev and OpenAI emit.

Context

Decision models are a small API category that answer structured questions — yes/no, multiple choice, or rubric-scoring — instead of generating free-form text. Three providers now compete: Jev (text and JSON only), OpenAI (text and images), and Google's Gemini Decision 1 (text, images, audio, video, and PDFs). Pricing is quoted per million input tokens, and Google and OpenAI charge nothing for output.

What to watch

• Watch for independent accuracy benchmarks comparing the three decision APIs and for Jev or OpenAI to add multimodal inputs or cut prices in response.

Related briefs

• OpenAIs GPT-6 Cyber can both break into systems and propose how to fix them
• The Iron(wood) Throne: Google's TPU Stakes Its Claim on Inference
• AI Text Watermarking May Not Be Able to Meet the EU’s New Rules
• Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.