OpenAI targets agent reliability with new Decisions API

The Decisions API allows the Luna model to select from a limited set of options, such as specific image categories or agent behaviors. This approach aims to maintain broad language and safety capabilities while slashing the latency and expense associated with standard, open-ended LLMs. TypeSafe CEO Diogo Almeida, a former OpenAI engineer, noted the development on social media, suggesting that the industry is shifting toward faster, intuitive decision-making models.

While OpenAI has positioned its new API as a limited preview, the underlying concept addresses a critical bottleneck in AI agent deployment. Currently, monitoring agents for misbehavior requires running secondary models at a significant compute cost. Cybersecurity expert Shapor Naghibzadeh demonstrated that using a Jev-like model for such oversight is drastically more efficient, estimating costs of less than $3 compared to over $370 when using a full-scale frontier LLM. By enabling granular, real-time review of every agentic action, these specialized decision models provide a practical path toward safer and more reliable autonomous systems.