OpenAI has opened its Decisions API, which narrows GPT-6 Luna to tasks with tightly bounded answers and returns a value an agent can branch on instead of generated text. The company says it cuts latency and cost on the small decisions that dominate agent loops and runs up to ten times faster than the Responses API.

  • Three answer formats: probability that a condition holds, selection from a supplied list, or a score on a scale.
  • Batching: multimodal context in, up to 400 questions per request.
  • Pricing: output tokens free, input at $0.10 per million tokens, or $0.01 per million with context caching.
  • Governance: processing localized in the US and EU, plus Zero Data Retention for enterprise customers.

The endpoint was shown at DevDay 2026 as a limited preview, so this announcement settles the details developers had been waiting on.

Related: OpenAI’s Decisions API Returns a Choice, Not a Paragraph, OpenAI Releases GPT-6 Sol and GPT-6 Luna

OpenAI Decisions API