OpenAI's Jev clone could help the frontier lab stop its swarming agents

OpenAI announced a new "Decisions API" at its Dev Day event, revealed by CEO Sam Altman. The API provides functionality similar to Jev, a model by TypeSafe AI explicitly designed for software automation. Jev is essentially a fast, cheap classifier built on an LLM that outputs probabilities for predefined choices. Altman described the API as a way to focus OpenAI's model on specific choices, making it "extremely fast" while maintaining image understanding, language support, and safety protections. TypeSafe CEO Diogo Almeida, a former OpenAI engineer who co-invented reinforcement learning, joked about the "clone wars" on X and suggested OpenAI's interest could be "a sign that building in a System One compatible way is the future." The underlying issue is that LLMs are comparatively slow and expensive for many software tasks, and decision models like Jev fill that gap. OpenAI's API is in limited preview, and other startups are developing similar products. A key application is monitoring and securing AI agents: following incidents where OpenAI's agents misbehaved on the open internet, the company uses a separate model to watch for bad actions at significant compute cost. Cybersecurity professional Shapor Naghibzadeh demonstrated at a hackathon that a model like Jev could perform this monitoring for just $2.94 per action versus $372 with a frontier LLM. This low cost means such monitoring could be run on every agent action, potentially improving agent reliability broadly. The article concludes that decision models have a promising future, particularly in agent monitoring and security.

(Source：TechCrunch)