Simon Willison releases llm plugin for OpenAI Decisions API

Summary

Simon Willison released an alpha version of a plugin (llm-openai-decisions 0.1a0) that enables the llm CLI tool to communicate with OpenAI's new Decisions API. This API mirrors the structure of Jev's decision endpoint but adds visual input support via the gpt-6-luna model.

Why it matters

The plugin offers developers a quick and familiar CLI path to OpenAI's structured decision endpoint, lowering the barrier to using LLM calls for prediction/selection/scoring in automation. Side-by-side comparison with Jev concretizes the emerging 'decision API' category. OpenAI's pricing is approximately 2.4 times Jev's input token rate.

Highlights

• gpt-6-luna accepts visual input; Jev's equivalent does not
• Both APIs charge only for input tokens (OpenAI $0.10/M, Jev $0.042/M)
• Willison asked GPT-6 Astra to read the documentation and generate code for this plugin by modeling it on the existing llm-typesafe plugin

Critical take

The plugin is still in 0.1a0 alpha, and the only demonstrated query is a single yes/no visual check. Therefore, real-world reliability for the three question types has not yet been proven. Additionally, despite having the same conceptual API structure, OpenAI's input token price is more than double Jev's. This will likely steer cost-sensitive developers toward the cheaper option.