OpenAI's Decisions API Hits Beta 10x Faster Typed Answers for Apps

What the Decisions API Does

OpenAI has released its Decisions API in public beta, introducing a dedicated endpoint that evaluates text and images and returns typed answers — not long-form text ¹. The only model currently supported is GPT-6 Luna, and OpenAI expects general availability in the coming weeks.

Instead of asking an LLM to generate prose and then parsing its output, developers send a request with shared evidence and a list of questions. Each question gets a unique name and a type, and the response returns an answers array keyed by those names ².

The goal is simple: cut the latency between feeding data into a model and getting a usable decision back.

Three Kinds of Answers

The API supports three question types:

Predicate checks whether a condition is true and returns a probability from 0 to 1. For example, it can assess whether a product photo shows a crack, tear, or dent.

Choice picks one value from a set of options you supply, returning per-option probabilities and a confidence score ². Use this for categories without an inherent order, like routing a support ticket to a department.

Score rates input against ordered levels, producing a probability-weighted average of the level indices. OpenAI uses a severity example: if an input has level probabilities of 0.1, 0.7, and 0.2, the resulting score is 1.1, landing between the second and third severity tiers ².

OpenAI draws a clear line: use the Decisions API for probabilities, choices, or scores; use Structured Outputs when you need to fill your own JSON schema or write explanations; and use function calling when a model must request a tool call with arguments ².

How Fast Is It?

OpenAI claims the Decisions API runs about 10× faster than its standard Responses API ¹. At DevDay, coverage put a single decision at roughly 150 milliseconds compared to about 1.6 seconds for a regular Luna call through the Responses API .

The company has not published independent accuracy or calibration data for the endpoint. The documentation advises developers to set their own thresholds using labeled examples from their application ¹.

Cost and Competition

Pricing for GPT-6 Luna through the Decisions endpoint is $0.10 per 1 million input tokens, with no output, cache-read, or cache-write charges ¹. Prompts above 272K tokens carry a 2× multiplier . The endpoint is available in the United States and Europe (EEA plus Switzerland), with Zero Data Retention and HIPAA compliance for eligible customers .

The closest rival is TypeSafe AI's Jev, launched in September 2026, which also returns typed values with calibrated probabilities. TypeSafe prices input at $0.042 per 1M tokens — roughly 2.4× cheaper than OpenAI — and reports end-to-end latency between 70 and 500 milliseconds ¹. However, OpenAI's edge includes image input support, compliance options, and broader beta access .

Key Takeaways

• The Decisions API returns probabilities, choices, and scores — not prose.
• Input costs $0.10 per 1M tokens; output is free.
• OpenAI claims about 10× faster responses than the Responses API.
• TypeSafe AI's Jev is cheaper but lacks image input and open beta access.
• GA is expected in the coming weeks.

Key facts

• Decisions API is in public beta as of October 6, 2026, powered by GPT-6 Luna
• Returns probabilities, choices, and scores in about 150ms — roughly 10x faster than the Responses API
• Input costs $0.10 per 1M tokens with no output charges
• Supports three question types: predicate, choice, and score
• TypeSafe Jev competes at $0.042 per 1M input tokens with 70–500ms latency

Sources

1. OpenAI Decisions API Hits Public Beta With 10x Faster Typed Answers — marktechpost.com
2. Decisions | OpenAI API
3. Decisions API is now available in Public Beta - Announcements - OpenAI ...
4. What is OpenAI's Decisions API? - Vercel — vercel.com
5. OpenAI Releases Decisions API in Public Beta, Powered by GPT-6 Luna — unite.ai