OpenAI has launched a new Decisions API that allows developers to build structured question-answering tools using their large language models. Simon Willison created a plugin called llm-openai-decisions to test the service, which is now available for installation via his llm command-line interface. The system uses the gpt-6-luna model to process inputs and return structured JSON responses based on specific query types.
This release matters because it introduces native image analysis capabilities alongside text processing. Users can upload files and ask questions about their visual content, receiving probabilistic answers in a standardised format. The pricing model charges only for input tokens, setting the cost at ten cents per million. This approach mirrors the earlier Jev system but adds visual understanding to the workflow.
* Input costs are ten cents per million tokens
* The model supports three distinct question types
* Responses return structured JSON with probability scores
