Google's Gemini Decisions API scores up to 1,000 yes/no and multiple-choice questions in one call
Google has launched the Gemini Decisions API, a new endpoint that answers closed questions instead of generating text. It returns calibrated probabilities in 150-250 ms and bills only input tokens, at $0.10 per million.
Google has launched the Gemini Decisions API, a new endpoint in the Gemini API built on a model called Gemini Decision 1 (gemini-decision-1). Instead of writing an answer token by token, it scores a fixed set of options in a single forward pass, returning probabilities for the options and one confidence score for each question's answer.
What it does
One request carries a shared input of up to 256K tokens, which can mix text, images, audio, video and PDFs, plus up to 1,000 independent questions about that input. Each question is one of three types:
- noul: the probability that a yes/no condition holds.
- choice: picks a winner from 2 to 255 labelled options and returns a probability for every option.
- score: grades the input against an ordered rubric and returns a weighted score.
Google quotes 150-250 ms end-to-end latency, staying nearly flat from 1 to 100 questions per request. Because every answer carries a confidence score, code can send ambiguous cases to a slower model or to a person.
What it costs
Input costs $0.10 per million tokens and output is free. The shared input is billed once per request, however many questions you ask, and each question adds its own instruction and criteria tokens. As a rough example, one million requests that each carry about 1,000 tokens of context come to about one billion input tokens, or about $100 for the context, plus the tokens for the questions themselves. There is also a no-cost tier in Google AI Studio for prototyping.
Where it fits
The endpoint targets the small decisions that agent pipelines make constantly: which tool to call, which model tier should take a query, whether content breaks a policy, or how good a retrieved passage is. Google says Gemini Notebook already uses it to route queries between model tiers, with up to 40% lower latency on everyday queries. For teams that today send these checks to a general chat model and parse the reply, it replaces free-form text with structured probabilities.
Two caveats come from Google's own announcement. The latency figures are for contexts up to 16K tokens, not the full 256K. And the benchmark results come from Google's own testing, against its own Gemini 3.5 Flash-Lite.
from google import genai
client = genai.Client(http_options={"api_version": "v1beta"})
result = client.decisions.create(
model="gemini-decision-1",
input=[{"type": "text", "text": "Refund request: the parcel arrived damaged."}],
questions={"route": {"type": "choice",
"instructions": "Which team should handle this?",
"criteria": {"billing": "Payments and refunds", "shipping": "Delivery problems"}}},
)Integrations are rolling out for OpenRouter, LangChain, the Vercel AI SDK, LiteLLM, Pydantic AI and others, alongside the google-genai SDKs.
Related on claudefox
- Gemini 4 Carbon is available to paying API developers, with launch pricing that doubles after November 30
- Gemini 4 Hydrogen costs $1.50/$7.50, and it only gets much cheaper than Carbon on December 1
- CleanYourTV: debloat an Android TV or Google TV over Wi-Fi with Claude Code and get an Apple TV-style home screen
- Google patched glitch tokens in the Gemma 4 E2B and E4B tokenizers: how to check your copy
Source: Gemini Decisions API: Fast, calibrated multimodal decisions (Google), Decisions API documentation (Google AI Studio)
This story first appeared in the claudefox daily briefing for November 12, 2026.