OpenAI announces Decisions API, which returns predefined answers with confidence scores, similar to TypeSafe's Jev, in limited preview with no pricing details
OpenAI has been expanding its developer platform from the Responses API for agent-building to remote MCP servers and built-in tools. Decisions adds a narrower interface to that stack: selecting from predefined outcomes rather than producing open-ended output.
The product also fits OpenAI's earlier investment in evaluation tooling for model performance, making confidence-scored choices more usable in workflows where developers can define the permitted actions.
First-order effects
Developers admitted to the limited preview can build applications around a bounded set of answers and receive a confidence score alongside the selection; pricing is not yet disclosed.
TypeSafe's Jev faces a directly comparable OpenAI offering, shifting the contest toward implementation quality and the economics of constrained decisions.
Second-order effects
Teams choosing between Decisions and Jev will need to benchmark decision quality, confidence behavior, and cost before standardizing on either API, particularly while OpenAI has not published pricing.
OpenAI can position constrained decisions alongside its existing agent-development API and tool integrations, giving customers a more specialized option for rules-bound workflow steps.
Third-order effects
If developers adopt the pattern, model APIs will increasingly be packaged around bounded, evaluable tasks rather than only general-purpose text generation.
Confidence-scored decision endpoints would reinforce demand for measurable AI procurement, where buyers assess useful task completion rather than model capability in isolation.
The trend: AI platforms are moving toward workflow-specific inference services that constrain model outputs and make automated decisions easier to evaluate.
Decisions API, for lightning fast constrained decision making powered by Luna. Supports visual inputs, and tuned to be able to make decisions in less than a few hundreds of milliseconds end to end.
curious if this is just an API wrapped on Luna? Luna is VERY cheap, but not Jev cheap Image inputs is great though, I've had it process images for $0.001
It was obvious, but f*ck that happened fast. Ok, so now I'll have to benchmark it, and I'll do so pretty fast. If anyone at @OpenAI can give me early/free access, to put it on JevBench, it would be appreciated. /cc @sama
New decisions API from OpenAI powered by giving Luna model a predefined set of options to choose from! Looks really good. Congrats to the Jev team for creating a new category too😉
OpenAI (and probably Anthropic and DeepMind) coming after Jev with their own decision model (aka System One Model) was an obvious and predictable outcome. However, this doesn't necessarily mean that Jev will be eliminated, because: 1) It was created by one of the former crack eng…