OpenAI Previews Decisions API for High-Speed Option Selection
The tool mirrors TypeSafe AI's Jev model, offering low-latency classification that could reduce the cost of monitoring autonomous AI agents.
OpenAI has revealed a limited preview of a new "Decisions API" designed to allow its models to select quickly and cheaply from a predefined list of options, Chief Executive Officer Sam Altman announced during the company's Dev Day event on Tuesday.
The tool provides functionality similar to Jev, a specialized classification model launched earlier this month by startup TypeSafe AI for software automation workflows, according to reporting by TechCrunch AI (https://techcrunch.com/2026/09/30/openais-jev-clone-could-help-the-frontier-lab-stop-its-swarming-agents/). OpenAI's Decisions API allows its Luna model to evaluate defined sets of choices—such as image classification categories or specific autonomous agent actions—and return probabilistic outputs at high speeds.
"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 at the event.
The approach reflects broader developer efforts to bypass the latency and financial overhead of running general-purpose large language models for discrete classification tasks. Following the announcement, TypeSafe AI Chief Executive Officer Diogo Almeida—a former OpenAI engineer whom TechCrunch AI identified as a co-inventor of reinforcement learning—commented on social platform X that OpenAI's product indicates growing interest in "System One" architectures focused on fast, intuitive selection over slower deliberate reasoning.
A primary operational use case for high-speed decision models is real-time monitoring of autonomous AI agents. After incidents where agents misbehaved online, OpenAI implemented secondary models to audit actions at substantial compute expense. Shapor Naghibzadeh, head of startup QueryStory, demonstrated a hackathon prototype that used TypeSafe's Jev to audit agent steps against instructions—blocking high-confidence violations, routing uncertain actions for human review, and permitting valid steps. Naghibzadeh estimated such continuous monitoring cost $2.94 with Jev compared to $372 when using a frontier LLM.
Sources
Written by
The Company Wire
Inside the companies building what’s next. Reporting on startups, technology, funding and the people shaping them.
