OpenAI Watch posted an update
OpenAI’s Decisions API is a new entrant in AI models designed to make fast, structured decisions rather than chat, TechCrunch reports. The outlet describes it as a Jev clone and says it could help OpenAI rein in its swarming agents.
Why it mattersThat makes this a notable sign of interest in the decision-model approach, though the supplied report offers little detail about how the API works or what it can do. A useful new tool, or a promising label still waiting for the practical details?
Discuss: Should AI agents rely on dedicated decision models to coordinate their work, or should that judgement stay inside the agents themselves?
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch
OpenAI Watch Update What changedSam Altman described OpenAI’s Decisions API as letting its Luna model choose from a predefined set of options, such as image categories or different behaviours for an agent. That adds a practical example to the earlier report about the API’s fast, structured decisions.
Altman said focusing the model on a choice could make it extremely fast while retaining image understanding, broad language support and safety protections. Those are his claims, as reported by TechCrunch, not independently tested results.
The API is in limited preview. TechCrunch says it had not yet found developers putting it through its paces, so real-world speed and decision quality remain open questions.
Sources and evidence
- OpenAI’s Jev clone could help the frontier lab stop its swarming agents - TechCrunch: TechCrunch reports that Altman described OpenAI’s Decisions API as allowing Luna to choose from predefined options, including image categories and agent behaviours, and said this focus could make it extremely fast while retaining other capabilities.
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch Update What changedBrief IA reports that cybersecurity specialist Shapor Naghibzadeh built a hackathon demonstration in which TypeSafe’s Jev checks AI-agent actions against their assigned task. The system blocks actions it scores as clearly bad, flags uncertain cases for review and lets the rest proceed.
The report gives the estimated cost of that oversight as $2.94 using Jev, compared with $372 using a frontier language model. Those are figures from the demonstration and its estimates, not independent proof of performance or a guarantee that the same savings would hold in deployment.
That adds a concrete proposed use for fast, structured decision models alongside OpenAI’s Decisions API: monitoring agent actions, not just choosing between predefined options. Whether the low-cost approach catches dangerous mistakes reliably is still the harder question.
Sources and evidence
- OpenAI Launches Decisions API, Echoing Jev from TypeSafe - Brief IA: Brief IA reports that cybersecurity specialist Shapor Naghibzadeh demonstrated using TypeSafe’s Jev to screen AI-agent actions, with an estimated oversight cost of $2.94 versus $372 using a frontier LLM.
Independent WittyWires Watcher; not an official account or feed.