Gemini Decisions API scores 1,000 multimodal questions in one pass

According to Google AI Blog, Google introduced the Gemini Decisions API, a new endpoint for fast, closed-set choices across text, images, audio, video and PDFs. The launch matters because it gives developers a way to route requests, grade content and apply policy checks without waiting for a model to generate free-form text.

Gemini Decisions API: Fast, calibrated multimodal decisions

What did Google launch with the Gemini Decisions API?

According to Google AI Blog, the Decisions API runs on Gemini Decision 1, or gemini-decision-1, through the Gemini API in Google AI Studio. The endpoint is designed for bounded choices such as tool selection, policy checks, model routing and context grading. Instead of asking a general model to write an answer, developers can send one multimodal input and attach up to 1,000 independent questions. The input can include text, images, audio, video and PDFs, with a context limit of up to 256K tokens.

Google says the API supports three decision types in one request. A yes/no question estimates the probability that a condition holds. A multiple-choice question selects from 2 to 255 options and returns normalized probabilities for candidates. An ordered rubric question grades an input across levels and returns a continuous weighted score plus per-level probabilities. Every answer includes a confidence score from 0 to 1, so code can route uncertain cases to a fallback or human reviewer.

Pricing is $0.10 per 1M input tokens with zero output tokens billed. The shared input is billed once per request regardless of question count, while each question adds only its instruction and criteria tokens. According to Google AI Blog, there are no output-token charges, variable thinking-token bills, or separate audio transcription and OCR costs.

How does Gemini Decision 1 make decisions faster than normal chat models?

The key difference is that Gemini Decision 1 does not generate answer text token by token. According to Google AI Blog, it scores candidate options in a single forward pass and returns probabilities and confidence scores. Google reports median model compute latency of 50-100 ms and end-to-end latency of 150-250 ms across multimodal inputs and contexts up to 16K tokens. Latency stays nearly flat from 1 to 100 questions per request.

That design also changes the cost profile. Because there are no output tokens to bill, Google says Gemini Decision 1 with the Decisions API is nearly 5x cheaper across the shared benchmark suite than Gemini 3.5 Flash-Lite with the Interactions API. It can process text, images, audio, video and PDFs directly in one pass. Developers can combine modalities, such as a call recording with account metadata or a scanned PDF claim with damage photos, and tune image or video resolution to control detail and token cost.

Google also reported benchmarks. According to Google AI Blog, across agent routing, reasoning, ranking and multimodal understanding, Gemini Decision 1 led the suite and had the lowest calibration error. That matters because the API returns a probability, not just a label, so code can decide whether to trust, retry or escalate.

What can developers build with the Decisions API?

Google gave several internal examples. Gemini Notebook uses the Decisions API as a real-time router to send incoming queries to the optimal model tier and thinking level. According to Google AI Blog, the API classifies intent in milliseconds, helping serve everyday queries with up to 40% lower latency while reserving deeper reasoning models for complex tasks.

Stitch's rapid canvas experiment uses the API to generate playable wireframes of a product flow in seconds and to apply edits made by voice or tap almost instantly. One click then hands the screens to full Stitch for deeper polish. CC from Google Labs uses the Decisions API to choose an emoji reaction from a palette of more than 200 options in a fraction of a second, so reactions appear without adding noticeable latency. Google Search uses the Decision API as a safety layer to detect adversarial behavior and content against adversarial evals with high efficacy within the latency envelope.

For prototyping, Google points to an API key in Google AI Studio, a no-cost tier, demo apps Doorstep Parcels, Speed Sketch and Find the Zoo Keeper, and Python, JavaScript and REST guides. Ecosystem integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode alongside the google-genai SDKs, with support coming soon to Gemini in Google Cloud.

How does this relate to Grafy?

The closest Grafy connection is not a one-to-one replacement, because Grafy is a creation workspace rather than a hosted decisions endpoint. Still, it is relevant to anyone working across multiple AI models. Grafy runs many cloud models side by side, and Grafy Chat lets users compare model behavior in one place. That can help build intuition about when a fast classifier-style decision is useful, though Grafy Chat is not the Gemini Decisions API.

For hands-on experimentation, Grafy Code is relevant for developers who want an AI coding studio to help draft and run API code. It can help explore the Python, JavaScript or REST patterns described by Google, then adapt the logic to a Grafy project. For document-heavy work, Reader can help users read, search and question PDFs and papers with AI. That is useful for understanding PDF inputs, but it does not provide the closed-set yes/no, choice or rubric scoring returned by the Decisions API.

Frequently asked questions

Does Gemini Decision 1 generate answer text token by token?

No. According to Google AI Blog, the model scores candidate options in a single forward pass and returns probabilities and confidence scores instead of decoding free-form answer text. That approach is central to the reported speed and to the pricing model, because there are no output tokens billed for the Decisions API.

Can the Decisions API evaluate images, audio, video and PDFs together?

Yes. Google says the endpoint accepts one multimodal input across text, images, audio, video and PDFs, up to 256K tokens, and allows up to 1,000 independent questions. The input can mix modalities, such as a call recording with account metadata or a scanned PDF claim with damage photos, and image or video resolution can be tuned.

Is the Decisions API limited to Google AI Studio?

No. According to Google AI Blog, ecosystem integrations are rolling out across OpenRouter, LangChain, Vercel, LiteLLM, Pydantic AI, Pipecat, LiveKit and OpenCode, alongside the google-genai SDKs. Support is also described as coming soon to Gemini in Google Cloud, giving developers broader deployment options beyond prototyping.

Source: Google AI Blog

---

Source: Google AI Blog — Gemini Decisions API: Fast, calibrated multimodal decisions