Google has launched the Gemini Decisions API, a multimodal service for scoring bounded choices such as tool selection and policy checks. Its practical pitch is speed and scale: one request can evaluate up to 1,000 questions, with the company giving end-to-end latency of 150 to 250 milliseconds.
Google DeepMind Watch analysis
What happened
The API is powered by Gemini Decision 1 and is available through the Gemini API in Google AI Studio. Developers can send text, images, audio, video or PDFs and ask yes-or-no questions, choose among options, or score against a rubric. Results include probabilities and confidence scores, rather than a generated answer in prose.
Google prices the service at $0.10 per million input tokens, with no charge for output tokens. The company says the shared input is billed once per request, even when it is used to answer multiple questions. Its announcement also describes Google products using the API for model routing, design iteration and other quick decisions.
Our top picks
- Up to 1,000 questions in one request
Score many checks against the same input, instead of making a separate model call for each one. - Text, images, audio, video and PDFs
A single request can combine different kinds of evidence, such as a call recording and account details. - Three decision types
Developers can request yes-or-no estimates, multiple-choice selections or scores against an ordered rubric. - Confidence scores for each answer
Applications can route uncertain cases to a fallback or a person, rather than treating every result as equally settled. - Input-only pricing
Google charges $0.10 per million input tokens and says output tokens are free.
Why it matters
This is a model endpoint built for the small, repeated decisions that steer larger systems: which model gets a request, whether content passes a rule, or which option an agent should choose. Handling many questions against shared input could make that kind of routing more practical, particularly where latency and cost matter.
The confidence scores are useful signals for software, not guarantees that a decision is correct. And the speed and benchmark advantages in Google’s announcement are company-reported figures, not independent results.
Our read
The interesting shift is from asking a model to explain itself to asking it to score a defined set of choices quickly. That could suit predictable workflows better than generating a paragraph and asking another system to interpret it. Developers should compare the costs and accuracy against their own tasks before putting those scores in charge; a tidy probability is still not a permission slip.
What to watch
- How the API performs on real developer workloads, beyond Google’s reported benchmarks.
- Whether the confidence scores reliably identify cases that need human review.
- How pricing and latency compare with conventional classifiers and general-purpose model calls.
Discussion spark: For routine decisions such as model routing or policy checks, would you trust a fast model score with a human fallback, or keep a person involved at every step?
Sources and evidence
- Gemini Decisions API: Fast, calibrated multimodal decisions (11 November 2026, 18:00 UTC)
not affiliated with, endorsed by, or operated by Google or Google DeepMind