OpenAI Opens Decisions API to All Users, Challenging TypeSafe’s Jev in Fast-Growing Decision Model Market

OpenAI made its Decisions API widely available to all customers on October 22, 2026, entering a market for fast, low-cost AI decision models that has already turned startup TypeSafe into a $7.6 billion company.

TypeSafe’s Jev model established the category by enabling deep learning models to act as rapid classifiers — given a set of options, they make quick choices and provide a confidence rating. Developers found these “System One” models could replace large language models in many workflows at significantly lower cost and higher speed. OpenAI’s product lead for APIs, Nikunj Handa, said his team saw Jev and recognized “a million use cases for this,” prompting a sprint to build a competing offering.

Rather than training a purpose-built decision model, OpenAI reconfigured its existing ChatGPT Luna model to reduce inference time and cost. The result cuts response times from one-to-three seconds down to 100–180 milliseconds. “Right now, it’s just plain old Luna that gets the job done,” Handa said. The API was first announced at OpenAI’s developer day on September 29, 2026, and the public release adds voice and visual modalities, along with geographically targeted inference to reduce latency globally.

Early enterprise customers including Box, Datadog, Ramp, and Speak tested the API ahead of launch. Speak CTO Andrew Hsu reported the tool cut the time between actions in the company’s AI language tutor from 1.4 seconds to 0.13 seconds. OpenAI also used the API internally to improve the responsiveness of its Dots voice assistant.

On public benchmarks such as the AIM System One Index and the Artificial Analysis Decision Index, OpenAI’s Decision Model ranks eighth and fourteenth, respectively, behind newer entrants including Liquid’s D1 Ultra and Base Lab’s Elephant-1. TypeSafe’s Jev remains the most queried decision model on OpenRouter. Handa pointed to multi-modal image classification as an area where Luna may hold an edge, and suggested further speed gains — potentially reaching 60 milliseconds — could open use cases such as per-frame video analysis.

The broader implication, as Handa described it, is an expanding set of use cases that become viable only as decision models grow faster and cheaper — a dynamic that mirrors the economic principle behind Jev’s namesake, which holds that greater efficiency in using a resource tends to increase its overall consumption.