OpenAI's Jev clone could help the frontier lab stop its swarming agents
News Source
β’Wed, 30 Sep 2026 19:00:57 +0000
π° What Happened
OpenAI showed a new Decisions API at its Dev Day event. It works like Jev, a model from TypeSafe AI that quickly picks between set choices. CEO Sam Altman said it lets OpenAI's Luna model choose from options, such as image types or agent actions. That keeps it fast and cheap while keeping language and image skills. TypeSafe's chief joked about a coming clone war.
π The Backstory
Jev is a fast classifier built on top of a large language model. Developers use it to make AI systems quicker and cheaper. TypeSafe says its edge is the synthetic data it creates. One coder showed that a Jev checker cost $2.94 to watch AI agent actions, compared with $372 using a big frontier model. Cheap checks like this could keep AI agents from doing harmful things.
π― Why It Matters
Cheaper AI checks could make agent tools safer and stop them from causing harm. That matters as more everyday software lets AI agents act on their own.
OpenAI showed a new Decisions API at its Dev Day event. It works like Jev, a model from TypeSafe AI that quickly picks between set choices. CEO Sam Altman said it lets OpenAI's Luna model choose from options, such as image types or agent actions. That keeps it fast and cheap while keeping language and image skills. TypeSafe's chief joked about a coming clone war.
Jev is a fast classifier built on top of a large language model. Developers use it to make AI systems quicker and cheaper. TypeSafe says its edge is the synthetic data it creates. One coder showed that a Jev checker cost $2.94 to watch AI agent actions, compared with $372 using a big frontier model. Cheap checks like this could keep AI agents from doing harmful things.
Cheaper AI checks could make agent tools safer and stop them from causing harm. That matters as more everyday software lets AI agents act on their own.