Key points
- Gemini Decision 1 is available through the Gemini API in Google AI Studio, with a no-cost tier for prototyping.
- One request can carry a single multimodal input of up to 256K tokens and as many as 1,000 independent questions.
- Google lists input at $0.10 per 1M tokens, no output charges, and says it is nearly 5x cheaper than Gemini 3.5 Flash-Lite.
- Integrations with OpenRouter, LangChain, Vercel and others are rolling out; Google Cloud support is listed as coming soon.
Why it matters for marketers: Marketers building AI workflows can use cheap, confidence-scored checks to sort, grade and screen content at scale, and send uncertain cases to a person.
Google introduced the Gemini Decisions API, an endpoint for bounded choices rather than open-ended writing. It runs on a new model, Gemini Decision 1, and handles text, images, audio, video and PDFs. Developers define yes/no checks, multiple-choice options or ordered rubrics, and the model returns a probability and a 0 to 1 confidence score for each answer instead of generating text.
For marketing and content teams, the useful part is the job type. Sorting incoming requests, grading drafts against a rubric, checking assets against brand or policy rules, and reviewing a call recording alongside account data are all closed-set decisions. Google says the shared input is billed once per request, so adding more questions costs little. Confidence scores let a workflow pass ambiguous items to a fallback or a human reviewer.
Google says its own products already use the API, including model routing in Gemini Notebook, quick prototyping in Stitch and a safety layer in Search. The Decoder notes that similar decision models have appeared in recent weeks, with offerings from Jev, OpenAI, Cloudflare and Microsoft, so this is a growing category rather than a one-off.
Next, watch how the integrations land in tools your team already uses, and when Google Cloud support arrives. Teams can test it now in Google AI Studio. The speed and cost figures come from Google, so run your own tests on real content before moving review work onto it.
Sources
This brief is our own summary. Read the full reporting at the source.