OpenAI made Decisions API available to all customers on Thursday, adding voice and image support to the public version. The broader rollout of the service, first announced at its developer day on September 29, also includes infrastructure changes aimed at reducing latency across regions worldwide.
Built on a reconfigured ChatGPT Luna
Decisions API is not based on a newly built, specially calibrated decision model. OpenAI API product lead Nikunj Handa said the team instead reconfigured ChatGPT Luna to accelerate inference and reduce costs.
Handa describes speed as the service’s main benefit: he said response times fell from one to three seconds to 100, 120 or 180 milliseconds.
Voice, images and use in the Dots assistant
Voice and visual modalities are included in the publicly available version. Handa highlighted image classification in particular, describing Luna as very strong at the task. He also cited better calibration without naming a model for comparison.
OpenAI also used Decisions API to improve the responsiveness of its own Dots voice assistant.
Speak reports shorter intervals between actions
Early customer trials included Box, Datadog, Ramp and Speak. Developers in those trials replaced OpenAI language models with Decisions API and achieved higher throughput.
Andrew Hsu, chief technology officer at Speak, which develops AI language tutors, reported a specific result. He said Decisions API helped the company reduce the time between actions from 1.4 seconds to 0.13 seconds.
OpenAI’s model trails Jev in both indices
TechCrunch reported that OpenAI’s decision model ranked eighth in the AIM System One Index, while TypeSafe’s latest Jev placed seventh. In the Artificial Analysis Decision Index, OpenAI’s model ranked fourteenth and Jev third.
