Sundar Pichai announces Gemini Decisions API for closed-set choices on X

Google CEO Sundar Pichai announced the Gemini Decisions API in an X post on 11 November. It is powered by a new model, Gemini Decision 1, in Google AI Studio. Everything below comes from that single post. OpenBeat has not seen a Google blog post or documentation page.

What happened

Pichai wrote on X that the API is built for closed-set choices rather than generating text token by token. He says it can score up to 1,000 questions in one pass. The inputs can be text, images, audio, video and PDFs.

He also said Google teams are already building with it. He shared a clip of Stitch Rapid designing a full product flow in seconds, shown at 1x speed.

Key details

All figures below are from Pichai's post, not from Google documentation.

• Launch: announced on 11 November, available through Google AI Studio.
• Model: Gemini Decision 1.
• Scale: up to 1,000 questions scored in a single pass.
• Input types: text, images, audio, video and PDFs.
• Decision types: three. Yes/no, multiple choice with up to 255 options, and rubric scoring.
• Price: $0.10 per 1M input tokens. Output tokens are billed at $0.
• Cost comparison: Pichai calls it roughly 5x cheaper than Flash-Lite.
• Quality claim: he says it leads on several benchmark evals and has the lowest calibration error.

What is not confirmed

The post does not name the benchmarks or say which models Gemini Decision 1 was compared against. It gives no scores. It does not say whether the API is a preview or generally available, and it lists no rate limits or regions. It does not explain how the 5x price comparison was worked out. No independent test is cited. The performance claims are Pichai's own.

Why it matters

Many developer tasks are closed-set choices: tagging, routing, moderation, grading against a rubric. A model that only returns a choice and bills nothing for output would change how you budget such jobs, if the claims hold. Treat the price and calibration figures as claims until Google publishes docs or outside tests appear.

For background on Google's other recent model launch, see Google's Gemini 4 Carbon public rollout.

FAQ

What is the Gemini Decisions API? According to Sundar Pichai's X post, it is an API powered by Gemini Decision 1 that returns closed-set choices instead of generated text. It handles yes/no, multiple choice and rubric scoring.

How much does the Gemini Decisions API cost? Pichai says $0.10 per 1M input tokens, with no charge for output tokens. He describes that as about 5x cheaper than Flash-Lite.

Are the benchmark claims independently verified? No. The post says the model leads several benchmark evals with the lowest calibration error. It names no benchmarks, and no independent results are cited.