OpenAI opens Decisions API to all customers

OpenAI made Decisions API generally available on Thursday, after first announcing it at its developer day on 29 September. The public release added voice and visual modes, and OpenAI said it had placed inference capacity in different regions to bring down delay.

Nikunj Handa, who leads APIs product work at OpenAI, said his team had seen TypeSafe’s Jev model and moved quickly to build a version of its own. He said the service came from reworking ChatGPT Luna rather than training a model built specifically for decisions.

Handa said the main benefit was speed, which he described as falling from one to three seconds to about 100, 120 or 180 milliseconds. He also said OpenAI could make the service faster still and use it for frame-by-frame video calls.

The article said early customers included Box, Datadog, Ramp and Speak, and that some used the API in place of OpenAI’s larger language models. Speak’s CTO, Andrew Hsu, said the time between actions fell from 1.4 seconds to 0.13 seconds.

The piece said it remained unclear what made a good decision model, and that benchmark results varied by workload. TypeSafe’s Diogo Almeida said he did not trust benchmarks, while the article noted that Jev led JevMan and that public indices put other models higher in some rankings.