The API is intended for applications that need to select a model, tool, or action in near real time rather than generate open-ended text.
Powered by GPT-6 Luna, it accepts text and image inputs. It returns one of three typed outputs: a predicate that estimates whether a statement is true, a choice from predefined options with confidence scores, or a score against a numeric range. OpenAI says the service can make decisions up to 10 times faster than GPT-6 Luna used through the Responses API.
Pricing is listed at $0.10 per million input tokens. OpenAI says only input tokens are billed; there are no output-token, cache-read, or cache-write charges. The announcement is dated October 6, 2026.
The public announcement does not provide independent accuracy results, calibration data for confidence scores, task-specific benchmarks, or service limits. It also does not establish whether the claimed speed improvement holds across particular workloads.