Felix Laurent
OpenAI's developer account said on oct 22 that the Decisions API is now generally available. The API lets apps and agents classify text and images, evaluate inputs and decide what happens next without generating a full model response.
What OpenAI announced
The announcement post says developers can add near real-time decisions to a workflow. OpenAI describes the result as faster, more cost-efficient apps and agents. The post gives no latency or price figures.
The API had been in public beta since oct 6, when OpenAI said it was open to all developers. That beta post described the use case as letting an app choose the right model, tool or action in near real-time.
The speed claim
In the beta post, OpenAI said the Decisions API makes decisions up to 10x faster than GPT-6 Luna through the Responses API. This is a vendor measurement. The general availability post does not repeat it.
The figure compares two different calls. One returns a decision, the other generates a model response. A routing step in an agent is the case where that gap could matter.
Price and outside testing
Cellcog, a secondary source, reports that the API charges $0.10 per million input tokens and nothing for output. It attributes that to OpenAI's guide. We did not open the pricing page, so treat the price as reported.
The same report says it has not seen an independent latency measurement. A separate early benchmark, cited by StackOne and dated oct 7, compared the API with a service called Jev. It found OpenAI 2x more expensive and 5 to 10% less accurate on one production task. That is a single task and a single report.
What to check before relying on it
The useful test is simple. Run the same classification or routing step through the Decisions API and through a normal model call. Compare latency, cost and accuracy on your own data. Until then, the faster and cheaper claim rests on OpenAI's statements.
