OpenAI announced a decision-making API. The company introduced Decisions API at Dev Day on September 30, 2026, describing it as a way to give the Luna model a predefined set of options to choose between. Its functionality resembles Jev, a model TypeSafe AI released earlier this month.
Sam Altman said the narrower task lets the model move faster without stripping away capabilities that developers expect from modern AI systems. “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.
The announcement puts OpenAI beside a model that has already attracted serious attention from developers. Jev is a new AI model built by TypeSafe AI, and it focuses on making decisions rather than generating text—the break from large language models associated with OpenAI and Anthropic is the point.
Jev Has Momentum Before OpenAI Arrives
Developers have been using Jev to augment LLMs, finding it faster and cheaper for decision-making tasks. The model’s launch video from mid-September 2026 has almost 40 million views on X, while a hackathon for Jev took place in San Francisco on September 29, 2026.
That attention has made Jev viral among developers, including those in San Francisco. Other startups are also rolling out models similar to Jev, turning a once-narrow product category into a small but busy race before OpenAI has explained how close its limited preview will be.
TypeSafe AI CEO Diogo Almeida, a former OpenAI engineer, joked about the beginning of the “clone wars” on X. He also said that building in a System One compatible way is the future, a reference to the decision-focused approach behind Jev.
The real dispute is not whether a model can choose between options. It is whether those choices contain enough intelligence to matter. Almeida framed the trade-off this way: “Fast and cheap is very easy, you know. 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.”
The Practical Case Is Agent Monitoring
Jev’s strongest use case may be monitoring AI agents. A model that checks whether an action matches its assigned task could make securing and reviewing agents affordable enough to run throughout an automated workflow, rather than only at expensive checkpoints.
Shapor Naghibzadeh, a cybersecurity professional leading QueryStory, built a demo that checks each agentic action against the task. The system blocks bad actions, flags others, and permits the rest—simple categories, but useful ones when an agent is operating without constant human supervision.
The cost difference is hard to ignore. Monitoring with Jev costs $2.94 versus $372 with a frontier LLM, according to the figures provided. That gap makes Jev arguably cheap enough to run on every agentic action, creating a review layer designed to improve agent reliability.
Decisions API may target the same pressure point, but its limits remain unknown because OpenAI released it as a limited preview. The announcement establishes similar functionality to Jev; it does not establish that the two models match in speed, price, intelligence, or monitoring performance.
Almeida says TypeSafe AI’s moat is the synthetic data it creates to generate statistically useful outputs. That claim points to the harder contest ahead: fast decision-making is easy to advertise, while producing reliable decisions across many tasks requires the data and intelligence to support them.
OpenAI’s move gives the category a larger stage and a familiar name. Jev supplied the viral launch, the developer experimentation, and the $2.94 monitoring example; Decisions API now brings Luna into the same conversation. The clone wars joke may be premature, but the market has already started copying the idea.
Based on
