OpenAI opens its Decisions API to all; Speak cut response time from 1.4s to 0.13s

OpenAI has made its Decisions API available to all customers, bringing a fast, low-cost model for choosing among options into its standard lineup. The model is a reconfigured version of its ChatGPT Luna model, tuned for speed and cost rather than built from scratch as a dedicated decision model. In OpenAI's account, response times dropped from one to three seconds to roughly 100 to 180 milliseconds.

The clearest customer figure comes from Speak, which builds AI language tutors. Its CTO, Andrew Hsu, said the API cut the time between actions from 1.4 seconds to 0.13 seconds, a reduction of about 90 percent. Box, Datadog and Ramp also tested the API in early trials, replacing larger LLM calls with a faster option. None of those companies has published a cost figure, so the savings case is still largely unquantified.

The benchmarks are less tidy. TypeSafe's Jev models and Liquid's D1 Ultra rank above OpenAI's offering on the public decision indices, and OpenAI's own model sits eighth and fourteenth on two of them. Its case rests more on multimodal support and calibration, which the company says are its strengths. For a business, the practical question is simple: does it hold up on your workload? Teams with high-volume, latency-sensitive choices, such as routing, moderation or voice interfaces, have the most reason to test it this quarter.