Simon Willison Builds Plugin for OpenAI Decisions API

Following OpenAI's recent DevDay 2026 announcements, developer Simon Willison has launched llm-openai-decisions 0.1a0, a new plugin designed for his command-line LLM utility. The tool integrates with OpenAI's newly released Jev-style Decisions API. To accelerate development, Willison used the GPT-6 Astra model to generate the plugin's code, drawing inspiration from his existing llm-typesafe plugin, which was built for the competing Jev platform.

The plugin leverages OpenAI's new gpt-6-luna decision model. Unlike Jev, which only processes text, gpt-6-luna supports both text and image inputs. The pricing structures of the two services also differ, though both charge exclusively for input tokens rather than output. OpenAI prices its Decisions API at 10 cents per million input tokens, whereas Jev offers a cheaper rate of 4.2 cents per million input tokens.

Conceptually, the OpenAI Decisions API mirrors Jev by supporting three distinct query types: yes/no questions, multiple-choice selections, and numerical scores. For example, a practitioner can pass an image of two pelicans to the gpt-6-luna model with a prompt asking if the image contains any mammals. The API returns a structured JSON response, such as a predicate evaluation showing a probability of 0.0.

For developers, this plugin simplifies the process of evaluating content and making automated classification decisions directly from the command line. By wrapping the Decisions API into a familiar toolset, practitioners can quickly prototype classification workflows, compare the cost-efficiency of OpenAI against Jev, and utilize multimodal inputs for structured decision-making tasks without writing custom API integration code.