Google launches Gemini Decisions API for multimodal choices
Google says its new Gemini Decisions API can score up to 1,000 multimodal questions in one request, returning probabilities and confidence scores for bounded choices.
Google has introduced the Gemini Decisions API, a new v1beta endpoint built around Gemini Decision 1 for making bounded choices rather than generating open-ended prose. The company says developers can use it for yes/no checks, multiple-choice decisions and rubric scoring across text, images, audio, video and PDFs.1
The announcement is aimed at jobs such as routing an agent to the right tool, checking a policy rule or grading incoming material against a fixed standard. One shared multimodal input can be assessed against up to 1,000 independent questions, according to Google, with each answer returning probabilities and a confidence score.
The API accepts inputs of up to 256K tokens, while Google says the model can handle contexts up to 16K tokens in the decision process. Multiple modalities can be combined in one request, such as a recording alongside account information or a scanned document alongside photographs. Developers can control image and video resolution to trade visual detail against token cost.
Google says Gemini Decision 1 returns results in 150–250 milliseconds end to end, with 50–100 milliseconds of that attributed to model computation. Pricing is listed at $0.10 per 1 million input tokens, with no output tokens billed. The company also claims the API is nearly five times cheaper than Gemini 3.5 Flash-Lite with the Interactions API on its shared benchmark suite. Those latency, calibration, performance and cost comparisons are Google’s claims, not independent testing.
The endpoint is available through the Gemini API and Google AI Studio in v1beta. Google’s announcement points developers to Python, JavaScript and REST documentation, but production readiness, quota limits and regional availability will need to be checked by teams before they build around it.
Why it matters
Most AI APIs are optimised for producing text. Gemini Decisions API targets a narrower but useful category: repeatable choices with a defined set of outcomes. That could make it easier to add confidence thresholds, human review and fallback models to automated workflows without asking a general-purpose model to explain every decision in prose.
Our opinion
This is a credible product direction because many practical AI tasks are classification or routing problems wearing a conversational interface. The value will depend less on the headline latency than on calibration, auditability and how reliably confidence scores identify cases that need a person. Google’s benchmark and cost claims are promising, but developers should test their own edge cases before treating a fast closed-set answer as a safe one.