OpenAI has made its Decisions API available to all customers. It’s the company’s answer to TypeSafe’s Jev, the model that created the market for fast, cheap “decision models” and helped make TypeSafe a $7.6 billion company. TechCrunch AI reports that the API went live for everyone on Thursday, after OpenAI first announced it at its developer day on September 29.
The backstory matters here. TypeSafe was founded by a former OpenAI researcher. Its Jev model showed developers that a lot of everyday AI work doesn’t need a full LLM. A super classifier that picks from a set of options and gives a confidence score often does the job faster and for far less money.
What OpenAI actually shipped
OpenAI didn’t build a purpose-made decision model. According to TechCrunch AI, Nikunj Handa, product lead for OpenAI’s APIs, said his team took ChatGPT Luna and reworked it so inference runs faster and costs less. “Right now, it’s just plain old Luna that gets the job done,” Handa said.
5 things worth knowing
- Speed is the headline. Handa called latency “the big unlock,” with response times dropping from one to three seconds down to about 100 to 180 milliseconds. That’s the gap between a noticeable pause and something that feels instant.
- New modalities in the public version. The general release adds voice and visual inputs. OpenAI also spread its inference resources across regions so customers get low latency wherever they are.
- Early customers saw real gains. Box, Datadog, Ramp, and Speak tested the API, mostly swapping out OpenAI’s LLMs to get more throughput. Speak, which makes AI language tutors, cut the time between actions from 1.4 seconds to 0.13 seconds, according to CTO Andrew Hsu. OpenAI also used the API to make its Dots voice assistant respond faster.
- Images could be OpenAI’s edge. Handa said Luna “is extremely powerful and has been showing better calibrations” for image classification. Customers are also asking for more flexibility around images as outputs.
- Even faster is the plan. Handa wants to get to around 60 milliseconds. At that speed, he suggested, you could call the API on every single frame of a video.
How it stacks up
On the public leaderboards, OpenAI isn’t winning. The AIM System One Index and the Artificial Analysis Decision Index both favor Liquid’s new D1 Ultra and Base Lab’s Elephant-1. The latest Jev ranks seventh and third on those indices. OpenAI’s model lands eighth and fourteenth.
TypeSafe still has the strongest adoption signal: its models are currently the most queried decision models on OpenRouter. TypeSafe’s newest Jev release adds image support and “smart-if statements.” It also tops JevMan, a benchmark built by Opper AI in which decision models play Pac-Man. Opper’s chief growth officer Felix Wunderlich said the game works as a test because of its “nonstop stream of quick, high-stakes decisions.”
The caveats
Nobody quite agrees on what makes a decision model good. The tradeoffs between intelligence, cost, calibration, and speed change from one workload to the next. TypeSafe CEO Diogo Almeida says he doesn’t trust benchmarks and tells developers to test models on their own tasks, even though TypeSafe just published its own rankings.
The article also doesn’t give pricing for the Decisions API. OpenAI pitches it as cheaper than its LLMs, but teams will need to run their own numbers.
Why this matters
What stands out to me is how OpenAI went about this. It didn’t spend months building a specialized model. It took one it already had and tuned it for speed, which tells you how much pressure a $7.6 billion startup can put on a frontier lab. It also points to a split in the market: big, slow models for reasoning, and quick, cheap ones for the thousands of small calls that happen inside real products.
Jev is named for the economist behind the Jevons paradox, the idea that using a resource more efficiently ends up making it more widely used. If that holds, cheaper decision calls won’t shrink AI budgets. They’ll show up in places nobody thought to use AI before. Handa calls it a “crazy set of use cases” that were hiding in plain sight. With Liquid, Base Lab, TypeSafe, and now OpenAI all in this race, expect prices to fall and speeds to climb quickly. TechCrunch AI has the full story.