At OpenAI's Dev Day on Tuesday, CEO Sam Altman revealed the new "Decisions API," a tool designed to provide high-speed and cost-effective classification for AI models.
Product LaunchesOpenAIサム・アルトマンDecisions API
OpenAI unveils "Decisions API" for high-speed, low-cost model classification
The API allows models, such as OpenAI's Luna, to choose from a predefined set of options—such as image categories or specific agent behaviors—rather than generating unstructured text. Altman stated that by focusing the model on these specific choices, the lab can maintain high capabilities in areas like image understanding, broad language support, and safety protections while significantly increasing processing speed.
The announcement follows the recent release of Jev by TypeSafe AI, a model specifically designed for software automation. TypeSafe CEO Diogo Almeida, a former OpenAI engineer, suggested that the interest in such tools indicates that building models in a "System One" compatible way—referring to fast, intuitive thinking—is the future of software integration.
Shapor Naghibzadeh, a cybersecurity professional, noted that specialized decision models could reduce the cost of monitoring AI agents. For instance, using a model like Jev to verify agent actions against assigned tasks can cost roughly $2.94, compared to $372 when using a frontier large language model.
Sources
- OpenAI’s Jev clone could help the frontier lab stop its swarming agents （TechCrunch AI, 2026-09-30）
- TypeSafe AI