Key points
- The Decisions API is now in public beta on the gpt-6-luna model.
- It operates roughly 10 times faster than the standard Responses API.
- Billing is $0.10 per million input tokens with zero charges for output.
OpenAI has launched its Decisions API in public beta, running exclusively on the gpt-6-luna model. According to MarkTechPost, this endpoint delivers typed answers approximately ten times faster than the existing Responses API. The service converts text and images into structured data points that code can immediately branch on, eliminating the need to parse natural language prose for labels.
By the numbers
- 10x: The speed improvement over the Responses API, per OpenAI.
- $0.10: Cost per million input tokens, with no output charges.
- 1,050,000: The token context window size for gpt-6-luna.
What the figures mean
The API supports three specific question types: predicate, choice, and score. A predicate checks conditions and returns a probability between zero and one. A choice selects a value from user-supplied options, providing per-option probabilities. A score rates input against ordered levels, calculating a probability-weighted average. OpenAI advises using this tool for decisions rather than Structured Outputs for JSON schemas or function calling for tool arguments.
Deployment is restricted to OpenAI-hosted servers via the POST /v1/decisions endpoint. No open weights are available, and self-hosting is not permitted. Each request includes shared evidence and a list of questions. The response returns an answers array keyed by the unique names assigned to those questions. This structure allows developers to handle complex logic without generating explanatory text.
OpenAI expects general availability in the coming weeks, according to MarkTechPost. Currently, only gpt-6-luna is supported, and the parameter count remains undisclosed. The company has not published independent accuracy or calibration data for the endpoint. Developers are advised to set thresholds using labeled examples from their own applications to ensure reliability, as no external evaluations have been released yet.
What should you do?
Security and IT teams should evaluate whether their applications require rapid, typed decisions rather than prose generation. If your workflow involves parsing model output for labels, this beta may reduce latency significantly. However, because no independent accuracy data exists, rigorous testing with labeled examples is essential before production use. Monitor the endpoint for updates on general availability and potential model expansions.
What to do and how to stay safe: OpenAI
- Test the beta endpoint with your specific data sets to verify accuracy thresholds.
- Compare latency metrics against your current Responses API implementations.
- Review budget impacts given the input-only pricing structure.
- Wait for general availability before integrating into critical production systems.
Step-by-step guide: Shadow AI Explained: 8 Questions Answered for Secure Operations
General security guidance from the BitShield News newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the cost of using the Decisions API?
It costs $0.10 per million input tokens, with no charges for output, cache reads, or cache writes.
Which model supports the Decisions API?
Only the gpt-6-luna model is currently supported, running exclusively on OpenAI-hosted servers.
How does the speed compare to other APIs?
OpenAI states the Decisions API runs about 10 times faster than the Responses API.
