The Decisions API returns structured typed outputs about ten times faster than the Responses API, with a simple per-token price and no output cost.
OpenAI has released its Decisions API to public beta, running on the GPT-6 Luna model. The API returns typed probabilities, choices, and scores, and is reported to be about ten times faster than the Responses API.
Pricing is set at $0.10 per million input tokens, with no output charges. The announcement provides no further details about limitations or usage, so the available information is limited to these performance and pricing claims.
OpenAI's case study shows how an AI agent, directed by an engineer, redesigned a production workflow to cut estimated model costs from $36 to $0.47 per run.
The retailer's OpenAI-powered Mylow tools have fielded tens of millions of customer and associate queries.
The choice between synchronous and asynchronous execution shapes latency, cost, and reliability for LLM-based agents in production.
The decision shows how generative AI is forcing scientific imaging contests to rethink their rules.