OpenAI's Decisions API turns text, images and JSON into probabilities, choices or scores at $0.10 per million input tokens. Here is where it fits and what businesses should test.
An AI feature does not need a long written answer every time. Many business applications only need to decide where a request should go, how risky it looks or which action should happen next.
OpenAI made its Decisions API generally available on 22 October 2026. It accepts text, images or structured data, then returns a probability, a choice from predefined options or a score against a rubric.
The price is $0.10 per million input tokens, with no output-token charge. Cached input costs $0.01 per million tokens. Developers can ask up to 400 questions in one request.
An e-commerce app could estimate whether a returned product looks damaged from a photo. A support system could route a complaint to billing, delivery or account security. An AI agent could choose whether to continue, retry or ask a person for help.
These jobs have bounded answers. That makes them easier to test than an open-ended chatbot response.
OpenAI says the dedicated API can run up to 10 times faster than GPT-6 Luna through its Responses API. Customer results vary by workload. Box reported about five times faster document routing across a test of 30 insurance certificates, while Sierra reported response times below 280 milliseconds at the 95th percentile for parts of its customer-service orchestration.
Those figures came from OpenAI and named customers, not an independent benchmark. A company should test its own data before budgeting around them.
Start with one repeated decision that already slows a workflow. Define the allowed answers and collect past examples with known outcomes. Measure accuracy, response time, cost and how often staff must correct the result.
Keep a human approval step for payments, account suspension, credit, medical decisions and other high-impact actions. A confidence score should guide review, not replace responsibility.
For African businesses handling high request volumes and tight cloud budgets, the value sits in repeated classification rather than open-ended chat. The low unit price can help, but integration, monitoring and review still cost money.
Before adding this to a custom web or mobile application, ask the development team to document the decision rules, fallback path, audit log and monthly volume. A small, measured workflow gives you a reliable cost and error baseline before wider deployment.
0 comments
Loading comments...