OpenAI introduced its Decisions API model to a broad audience, marking its entrance into the fast, affordable intelligence sector recently responsible for turning TypeSafe—a startup founded by a former frontier lab researcher—into a $7.6 billion company.
TypeSafe’s Jev model established a category for deep learning systems acting as super classifiers: given options, they rapidly make choices and provide confidence ratings. Developers found these "System One" models could replace LLMs in workflows, operating faster and cheaper.
Nikunj Handa, OpenAI's API product lead, noted his team saw "a million use cases" after seeing Jev. They reconfigured ChatGPT Luna to speed up inference and lower costs rather than building a specific decision model.
Handa highlighted that the Decisions API's main benefit is speed, reducing latency from one to three seconds to 100 to 180 milliseconds, stating, "The big unlock with Decisions API has been the speed, shrinking things from one to three seconds to 100, 120, 180 milliseconds."
Announced at the September 29 developer day, the API is now live. OpenAI added voice and visual modalities and optimized global latency by targeting inference resources geographically.
Trials with Box, Datadog, Ramp, and Speak showed engineers replacing LLMs with higher throughput. Speak CTO Andrew Hsu noted the API cut action time from 1.4 seconds to .13 seconds, and OpenAI used it to improve the Dots voice assistant.
It remains unclear what defines a "good" decision model, as optimization varies. TypeSafe CEO Diogo Almeida distrusts benchmarks but released rankings. Public benchmarks like the AIM System One Index favor Liquid's D1 Ultra and Base Lab's Elephant-1, though TypeSafe's Jev ranks 7th/3rd, and OpenAI's model ranks 8th/14th.
TypeSafe's model is currently the most queried on OpenRouter. The JevMan benchmark, where models play Pac-Man, shows Jev topping the rankings, which Felix Wunderlich of Opper AI calls interesting due to the need for quick, high-stakes decisions.
Handa suggests OpenAI's model has an advantage in multi-modality, particularly image classification, where "Luna is extremely powerful." He also mentioned future goals to reach 60 milliseconds for video frame processing.
As these tools become cheaper and faster, they may live up to Jev's namesake—a commodity economist who predicted ubiquity—as Handa noted these "crazy set of use cases" were hiding in plain sight.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.