OpenAI revealed a new product called the Decisions API at its Dev Day event on Tuesday. CEO Sam Altman mentioned the tool during the event.
The API appears to work much like Jev, a model released by TypeSafe AI earlier this month. Jev is built for software automation.
Jev acts like a powerful classifier built on a large language model. Developers give it a set of choices, and it returns probabilities for each one quickly and at low cost.
How the Decisions API Works
Altman said the Decisions API gives OpenAI's Luna model a fixed set of options to choose from. These could be categories for sorting an image or different behaviors for an AI agent.
"By focusing the model on that choice, we can make it extremely fast while keeping capabilities like image understanding, broad language support, and safety protections," Altman said.
OpenAI released the API as a limited preview. It is not yet clear how closely it matches Jev, and few developers have tested it publicly so far.
Still, posts on X show that developers are interested. Other startups are also rolling out similar decision models.
TypeSafe AI Responds
TypeSafe did not answer TechCrunch's questions about OpenAI's new product. CEO Diogo Almeida, a former OpenAI engineer, joked on X about the start of the "clone wars."
He also said OpenAI's interest could be "a sign...that building in a System One compatible way is the future." TypeSafe uses "System One" to describe fast, intuitive thinking and "System 2" for slower, careful reasoning.
Developers have used Jev alongside large language models. Some have reported that this makes their software faster and cheaper.
Almeida says his company's edge is the synthetic data it creates to produce useful results. "Fast and cheap is very easy, you know," he told TechCrunch last week.
"If you want it really fast and cheap, use dice, right? Intelligence is the hard part, and my North Star is always pushing the intelligence-per-dollar Pareto curve," he said.
One possible use for these models is watching over AI agents. OpenAI recently added security measures after several incidents in which its agents misbehaved on the open internet.
One of those measures uses a separate model to watch for bad actions. OpenAI has said this comes at a high computing cost.
Shapor Naghibzadeh, a cybersecurity professional who leads the startup QueryStory, believes a model like Jev could do the same job for less. He built a demo at a hackathon held last weekend.
His demo uses Jev to check each agent action against its assigned task. It blocks actions it is confident are bad, flags others for review, and allows the rest.
In theory, this kind of monitoring could have stopped the Hugging Face incident. It costs $2.94 with Jev, compared with $372 using a frontier large language model.
