Google launches Gemini Decisions API

Google's Gemini Decisions API answers up to 1,000 yes/no, choice or rubric questions per call at $0.10 per million input tokens, with no output fee.
What it means for founders
- Pack your checks into one call. Google bills the shared input once per request, however many questions ride on it (each adds only its own instruction tokens), up to 1,000. Run a dozen policy or routing checks on one ticket or clip in a single request. There is no context caching, so resent input is billed in full. OpenAI charges $0.01 per million for cached input.
- Same list price as OpenAI. Google's $0.10 input rate and zero output charge equal what OpenAI charges for its Decisions API. Test both on your own data. OpenAI caps a call at 400 questions; Google allows 1,000 per call.
- Keep the integration portable. On November 10, Hugging Face proposed a common interface for decision models, built on the same three question types Google uses (noul, choice and score), with support in Transformers and llama.cpp. Wrap these calls in your own function so swapping providers stays cheap.
- Put a person on money decisions. Google's docs warn that user-supplied text may lobby for the label it wants, and advise a human or second check before actions such as payouts.
How the Gemini Decisions API works
Each request carries one shared input of text, images, audio, video or PDFs, plus named questions, all within a 262,144-token context window. A question takes one of three forms: a yes/no probability (capped at 0.01 and 0.99), a pick from 2 to 255 options, or a score against an ordered rubric. Each answer adds a 0 to 1 confidence value, which lets you hand unclear cases to a fallback or a reviewer.
The model scores every option in one forward pass. For inputs up to 16K tokens Google quotes a median of 150 to 250 milliseconds end to end, with little change between 1 and 100 questions. On its own benchmark suite, Google puts the cost at nearly five times below Gemini 3.5 Flash-Lite run through the Interactions API.
Gemini Decisions API pricing and limits
AI Studio offers a no-cost tier for testing, but Google's pricing page says it uses free tier data to improve its products, unlike paid tier data. Free accounts get 200 requests a minute and 10,000 a day, and the top paid tier allows 15,000 a minute.
It runs on the v1beta Gemini Developer API only, without Batch, Flex or Priority inference. Integrations with OpenRouter, LangChain, Vercel, LiteLLM and others are rolling out, and Google says Gemini in Google Cloud support is coming soon.
Google says Gemini Notebook uses it to pick a model tier and thinking level per query, cutting latency on everyday questions by up to 40%, and that Search uses it to screen MCP tools for adversarial behavior.
How Google says it stacks up against Jev and OpenAI
Google says its model beat TypeSafe AI's Jev and OpenAI's Decisions API on seven of nine benchmarks it ran, SiliconANGLE reports, though Jev proved slightly more cost-efficient. Google's post also claims the lowest calibration error.
What we don't know yet
- How those benchmark wins hold up outside Google's own tests.
- When the API reaches Google Cloud, and whether it will offer regional processing (the docs we read are silent on it).
- Whether caching or Batch support will follow.
The story
Google launched the Gemini Decisions API on November 11, an endpoint on the Gemini developer platform that returns probabilities for closed questions in place of generated text. It runs on a new model, Gemini Decision 1, and costs $0.10 per million tokens of input, with output free, according to Google's announcement.
Sources
Primary sources
Tools in this story
We may earn a commission if you sign up through our links. It never affects our ratings or which stories we cover.
Google's assistant, built into Workspace and Android
More in Products & Launches
- ElevenLabs launches ElevenAI note-taker for teams

For founders: Teams picking a note-taker: the free window makes a pilot cheap, but ElevenLabs lists no price after it ends.
TechCrunch · 5d ago - OpenAI launches ChatGPT Wallet for agent purchasesChatGPT Wallet lets OpenAI's dots agents buy for you with tokenized cards and Face ID or fingerprint approval.OpenAI · 5d ago

- Anthropic launches Claude Sheets and WhiteboardsAnthropic put Claude Sheets and Whiteboards in beta and made Claude Data generally available. Anthropic lists Meetings as coming soon.TechCrunch · 8d ago
- OpenAI lets businesses claim profiles in ChatGPT

For founders: Claim the profile now. It is free, and Shomair says a profile "should not have an impact" on how often a business appears, though clearer data could help…
OpenAI · 8d ago - Adobe will open Creative Cloud apps to AI agents

For founders: An opening, on Adobe's terms. For agent or workflow tools aimed at creative teams, editing real Photoshop and Premiere files is a real hook.
TechCrunch · 8d ago