Key points
- OpenAI launched the Decisions API in public beta on the GPT-6 Luna model.
- The new interface returns typed probabilities and scores approximately ten times faster than the Responses API.
- Billing is set at $0.10 per million input tokens, with no charges for output tokens.
OpenAI has moved its Decisions API into public beta, marking a shift in how developers can access structured data from its latest models. According to MarkTechPost, the service is now available on the GPT-6 Luna model, offering a distinct alternative to the company’s standard Responses API for applications requiring specific data types.
By the numbers
- 10x faster response times compared to the standard Responses API.
- $0.10 cost per one million input tokens for using the Decisions API.
- Zero cost per million output tokens, as billing is input-only.
What the figures mean
The speed increase is the most immediate technical change for engineering teams. MarkTechPost reports that the Decisions API returns typed probabilities, choices, and scores about ten times faster than the Responses API. This performance gap matters for applications that need rapid, structured decisions rather than open-ended text generation.
The pricing structure differs significantly from standard usage models. Developers pay $0.10 for every million input tokens sent to the GPT-6 Luna model via this endpoint. There are no charges for output tokens, which could reduce costs for workflows that generate large amounts of structured data from concise prompts.
That suggests a strategic pivot toward efficiency for specific use cases. By separating structured decision-making from general text generation, OpenAI allows teams to optimize for speed and cost in parallel. The absence of output billing removes a variable expense for high-volume data extraction tasks.
What to watch
Security operations teams should monitor how this speed and cost structure influences application architecture. Faster, cheaper structured outputs may lead to higher query volumes, increasing the attack surface for prompt injection or data leakage. Teams using GPT-6 Luna should review their input validation strategies, as the lower cost per token might encourage less rigorous filtering.
Background: Input validation
Input validation is the process of checking that data meets strict format, type, length, and range requirements before processing. It prevents injection attacks and logic bugs by ensuring the system only handles expected values. This defense operates at the application layer, complementing network controls and secure coding practices.
Read the full guide: Input Validation: How to Stop Bad Data at the Border
What to do and how to stay safe: OpenAI
- Review application logs to ensure input validation rules remain strict despite lower API costs.
- Monitor token usage metrics to detect anomalies that could indicate prompt injection attempts.
- Verify that structured outputs from the Decisions API are sanitized before entering internal systems.
- Update rate-limiting configurations to account for the potential increase in query volume due to faster response times.
Step-by-step guide: Detect Shadow AI: Network Logs, Process Trees and Hidden Signals
General security guidance from the Sandbox Gazette newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What model powers the new Decisions API?
The Decisions API runs on the GPT-6 Luna model, according to MarkTechPost.
How much does the Decisions API cost?
It costs $0.10 per million input tokens, with no charges for output tokens.
Is the Decisions API faster than the Responses API?
Yes, it returns typed probabilities and scores about ten times faster than the Responses API.
