By Hamza Faisal

OpenAI has opened its Decisions API to every customer, stepping squarely into the market for fast, cheap AI decision-making that turned startup TypeSafe's Jev model into a $7.6 billion company within weeks of its launch. Announced at the company's developer day on September 29, the API became generally available on Thursday, October 22, with new voice and visual support in the public release.

The big unlock is speed

Instead of training a dedicated decision model from scratch, OpenAI's API team took ChatGPT Luna and reconfigured it to run faster and cheaper. Product lead Nikunj Handa told TechCrunch that response times fell from one to three seconds to 100 to 180 milliseconds — fast enough to swap out full LLM calls in routing and selection steps. "The big unlock with Decisions API has been the speed, shrinking things from one to three seconds to 100, 120, 180 milliseconds," Handa said. He added that his colleagues took one look at Jev's traction and concluded: "we have like a million use cases for this." OpenAI also routed inference geographically to keep latency low for users around the world.

A $7.6 billion category to chase

TypeSafe's Jev pioneered a new class of "decision models" — sometimes called System One models — that take a set of options, quickly pick one, and attach a confidence rating to the choice. Developers rapidly adopted them as drop-in replacements for large language models in classification and triage workflows, where they run far faster and cost dramatically less. TypeSafe was founded by a former OpenAI researcher, and Jev's success made it a $7.6 billion company almost overnight — a signal OpenAI clearly didn't want to miss.

Early results and what comes next

Early testers including Box, Datadog, Ramp and Speak swapped the API in for OpenAI's own large models to boost throughput. Speak, which builds AI language tutors, cut the delay between actions from 1.4 seconds to roughly 0.13 seconds, according to CTO Andrew Hsu. OpenAI has also used the technology internally to sharpen the responsiveness of its own Dots voice assistant. Handa's next target is a 60-millisecond call — enough, he said, for "an API call for every single frame in a video." The launch underscores how quickly frontier labs are moving to commoditize the inference layer beneath AI agents — speed, not just intelligence, is now the battleground.

By Hamza Faisal