Decisions API is now available in Public Beta - Announcements - OpenAI Developer Community
Short summary
OpenAI released the Decisions API in public beta, which makes classification and decision-making tasks roughly 10 times faster than their Responses API. The tradeoff is higher cost per request and currently no token caching support, which some developers find economically unattractive for certain use cases.
- Decisions API processes text and image inputs to return typed answers (probabilities, choices from sets, rubric scores) about 10x faster than the Responses API, with median response times around 0.3-0.8 seconds on typical connections
- Accuracy is comparable to Luna (the model used in Responses API) but slightly faster and more variable in performance, though the practical difference is negligible
- No token caching is currently available, making the API more expensive per request than alternatives for repeated classification tasks—a limitation developers hope gets addressed soon
Full summary
OpenAI released the Decisions API in public beta, a tool built for fast classification and decision-making tasks. It returns answers about 10 times faster than the existing Responses API by evaluating text, images, or both and outputting typed results—probabilities, choices from fixed sets, or scores against rubrics. In real-world tests, the median response time was around 0.3 seconds compared to Luna's 1.6 seconds. The tradeoff is cost: Luna runs 2–3 times more expensive, but it's slightly more accurate and better calibrated, though the difference doesn't matter much in practice.
The API doesn't currently support caching, which is a significant limitation for classification workflows. One developer noted that without caching, switching from Luna to Decisions becomes economically unattractive since you can't reuse cached input tokens to reduce costs. OpenAI staff acknowledged this gap and suggested caching might arrive in future updates, but there's no timeline. For real-time applications like game AI, the speed advantage is genuine—one user reported 0.8-second responses for image inputs over a slow 3 Mbps connection.
The main uncertainty centers on decision quality when multiple valid options exist. Developers are asking whether Decisions API will reliably pick the optimal choice when several paths are equally plausible, or if it just defaults to the statistically most likely outcome. The community discussion hints that the API optimizes for the highest-probability result, which works fine for straightforward classification but might miss nuance in complex scenarios where multiple "right" answers exist.