OpenAI's Decisions API Goes Live, Taking Direct Aim at Jev

OpenAI opened its Decisions API to all customers Thursday, cutting response times from one to three seconds down to 100-180 milliseconds.

The launch puts OpenAI into a category TypeSafe's Jev model created: deep learning models that act as super classifiers. Given a set of options, they quickly make choices and offer a confidence rating in their decision. Developers found these "System One" models could replace LLMs in a variety of workflows, running more quickly and at dramatically lower cost. TypeSafe, founded by a former researcher at a frontier lab, is now a $7.6 billion company on the back of it.

A repurposed Luna, not a purpose-built model

OpenAI did not build a specially calibrated decision model. Nikunj Handa, product lead for OpenAI's APIs, told TechCrunch his team took ChatGPT Luna and reconfigured it to speed up inference and lower cost.

"In the future, we may sort of like have specific training to make our models like even better in these ways. But right now, it's just plain old Luna that gets the job done." — Nikunj Handa, OpenAI

That matters when sizing up the API: the speed comes from serving changes, not from a model trained for classification. Handa said colleagues saw Jev and said "wow, we have like a million use cases for this," and his team sprinted to ship its own version.

The speed numbers

Announced at OpenAI's September 29 developer day, the API reached general availability Thursday. The public version added voice and visual modalities, and OpenAI targeted inference resources geographically to lower latency by region.

Box, Datadog, Ramp and Speak ran initial trials, swapping out OpenAI's LLMs for higher throughput. Andrew Hsu, CTO at language-tutor company Speak, said the Decisions API cut the time between actions from 1.4 seconds to 0.13 seconds. OpenAI also used the product to improve the responsiveness of its Dots voice assistant.

Where it ranks against Jev

Public benchmarks do not favor OpenAI. The AIM System One Index and the Artificial Analysis Decision Index both tend to favor D1 Ultra, a decision model Liquid released last week, and Base Lab's Elephant-1.

Model	AIM System One Index	Artificial Analysis Decision Index
Latest Jev (TypeSafe)	7th	3rd
OpenAI Decision Model	8th	14th

TypeSafe's offerings are, for now, the most widely queried on OpenRouter. Its latest Jev release added image support and smart-if statements, and tops JevMan, a benchmark where decision models play PacMan. Felix Wunderlich, chief growth officer at Opper AI, which built the benchmark, said the game "is an interesting test because of its nonstop stream of quick, high-stakes decisions."

Is it worth switching?

The rankings argue against OpenAI on text classification, but a benchmark position is a weak signal in this category. TechCrunch's reporting notes it is not clear what makes a good decision model beyond "it works for you," since the tradeoff across intelligence, cost, calibration and speed appears to differ by workload. TypeSafe CEO Diogo Almeida distrusts benchmarks and urges trialing models on your own work — though TypeSafe itself published rankings last week alongside its newest Jev.

The clearer OpenAI advantage is multimodality. Handa said that for image classification, Luna "is extremely powerful and has been showing better calibrations," and that customers want the API to be more flexible about using images as outputs too. If your decision step is visual, that is the reason to run the comparison. If it is pure text, Jev and D1 Ultra deserve the first trial.

What to watch

Handa floated pushing latency lower still, asking whether the team could "get this to something like you know 60 milliseconds, and then you could imagine like an API call for every single frame in like a video." He framed it as a question, not a roadmap commitment — but that threshold is the one to track, because per-frame decisioning is a different product shape than per-request classification.