OpenAI Decisions API
ActiveOpenAI's typed-answer endpoint: probabilities, choices and scores from text or images
The Decisions API turns text or images into typed probability, choice or score answers for $0.10 per 1M input tokens with no output charge, and OpenAI says it is about 10x faster than the Responses API. Conditional: it is in public beta, supports only gpt-6-luna, and GA is only expected in the coming weeks. The speed claim is OpenAI's own. It suits teams routing or classifying at volume, not anyone who needs generated text.
Every call is a gpt-6-luna model evaluation returning typed answers; there is no non-model path in the product.
Is it right for you?
Good for
- Classifying, routing and prioritizing text, images or JSON where you want a probability, a choice or a score back instead of free text to parse
- High-volume decision calls on cost: $0.10 per 1M uncached input tokens, discounted cache hits and no output-token charge
- Regulated workloads: Zero Data Retention and HIPAA use for eligible customers, with data residency in all supported API regions
- Routing and filtering on thresholds you set: choice and score answers return a probabilities array and a confidence value, and a request can carry up to 400 questions
Not good for
- Generating or editing text, or running multi-step tool use: it returns typed answers to the questions you ask
- Choosing your model: gpt-6-luna is the only supported model
- Exact cost planning in other regions or on long inputs: regional premiums and long-context multipliers apply, and gpt-6-luna input is $0.20 per 1M above 272K tokens
Our experience
Pricing
Paid- No output-token charge
- Cache hits discounted at standard gpt-6-luna caching rates
- Regional processing premiums and long-context multipliers apply
- gpt-6-luna only, generally available since October 22, 2026
No verdict changes yet
The clock starts day one. Changes land here as our verdict evolves.
Sources
Verification log
- Status· Updated
Automated agent
Decisions API went generally available on Oct 22 2026 (stored: public beta, GA expected in coming weeks). GA raised the per-request limit from 200 to 400 questions, added prompt-cache discounts for shared input, input_json and images in question instructions. Oct 15: native GPT-Live integration. Uncached input still $0.10 per 1M, no output charge, gpt-6-luna still the only model. Stored claim of no cache-read charge replaced: cache hits are discounted at gpt-6-luna caching rates. Data residency now in all supported API regions. Status set to active; description, strengths, audiences, tier and sources refreshed. Verdict summary still says public beta and cannot be written by tool_update; restated summary staged for publish.
- Pricing· No changes
Automated agent
Want this running in your business?
We tested this tool and we build with tools like it every day. Tell us the workflow and we will set it up, integrate it and hand it over working.
The conversation runs on Gnosari, one of the tools in this directory. A real conversation, not a sales script.