OpenAI has opened its Decisions API model to all users for the first time, stepping into the fast, inexpensive intelligence market that recently turned TypeSafe—a startup founded by a former frontier‑lab researcher—into a $7.6 billion company. TypeSafe’s Jev model pioneered a class of deep‑learning systems that act as “super classifiers”: given a list of options, they can swiftly select one and attach a confidence score to the choice. Developers quickly realized that these decision‑or “System One”—models could supplant large language models in many workflows, delivering higher speed at dramatically lower cost.
Nikunj Handa, product lead for OpenAI’s APIs, told TechCrunch that his team saw Jev and exclaimed, “wow, we have like a million use cases for this.” In response, OpenAI’s engineers rushed to deliver their own version. Rather than engineering a purpose‑built decision model, they took the existing ChatGPT Luna system and re‑engineered it for faster inference and reduced expense. “The big unlock with Decisions API has been the speed, shrinking things from one to three seconds to 100, 120, 180 milliseconds,” Handa said. “In the future we may add specific training to make our models even better in these ways, but right now it’s plain old Luna that gets the job done.”
Announced at OpenAI’s developer day on September 29, the Decisions API became generally available on Thursday. The public release added voice and visual modalities and aimed to lower latency worldwide by locating inference resources closer to users. Early trials with companies such as Box, Datadog, Ramp and Speak showed engineers swapping OpenAI’s standard LLMs for the new API to achieve higher throughput. Andrew Hsu, chief technology officer at Speak, which builds AI‑driven language tutors, reported that the Decisions API cut the interval between actions from 1.4 seconds to 0.13 seconds. OpenAI also used the service to make its Dots voice assistant more responsive.
What constitutes a “good” decision model remains fuzzy beyond the simple test of “it works for you.” Trade‑offs among intelligence, cost, calibration and speed vary by workload. TypeSafe CEO Diogo Almeida is skeptical of benchmark rankings and encourages users to test models in their own contexts, though his company released its own comparative scores last week alongside the latest Jev model, which now supports images and smart‑if statements.
Public benchmarks such as the AIM System One Index and the Artificial Analysis Decision Index currently favor D1 Ultra—a new decision model from Liquid—and Base Lab’s Elephant‑1. The newest Jev model placed seventh and third on those lists, while OpenAI’s Decision Model fell to eighth and fourteenth, respectively. Nevertheless, TypeSafe’s offerings remain the most frequently queried on OpenRouter.
A lighter‑hearted competition, JevMan, pits decision models against the classic Pac‑Man game. The latest Jev model leads that leaderboard. Felix Wunderlich, chief growth officer at Opper AI, which created the benchmark, said the game “is an interesting test because of its nonstop stream of quick, high‑stakes decisions. At every junction the model must decide where to go in a few hundred milliseconds.”
OpenAI’s model may have an edge in multimodal tasks. Handa noted that for image classification “Luna is extremely powerful and has been showing better calibrations,” and that customers are asking for the Decisions API to handle images as outputs more flexibly. He also hinted at future speed gains that could enable new scenarios: “Can we get this to something like 60 milliseconds, and then you could imagine an API call for every single frame in a video?”
As decision‑model services become faster and cheaper, they may live up to the namesake of Jev, an economist who argued that efficient resource use makes technology more ubiquitous. Handa envisions an expanding “crazy set of use cases” that were hidden in plain sight until decision models brought them to light.