Intel / article

llm-openai-decisions 0.1a0

Source: simonwillison.net
Date: Published Oct 6, 2026

Context

Simon Willison released an alpha plugin for his command-line tool that calls OpenAI's Decisions API, a service that answers a question about text or an image with a typed result and a probability instead of prose. He had GPT-6 Astra read the API documentation and write the plugin, modeled on an earlier one he built for Jev, the comparable model from TypeSafe AI.

His comparison is the useful part. OpenAI's gpt-6-luna decision model accepts images as well as text, while Jev, by his account, takes text. Both charge only for input: 10 cents per million input for OpenAI and 4.2 cents for Jev. Both support three question types, yes or no, a choice from options, and a score.

His example asks whether an attached image (a file named two-pelicans.jpg) contains any mammals and gets back a predicate result with a probability of 0.0. That shows the output shape from one test, not how accurate the model is. The release is version 0.1a0, an early alpha.

Key takeaways · AI-distilled

Unlike Jev, OpenAI's gpt-6-luna decision model accepts image input as well as text, Willison notes.
Both services bill only input tokens and charge nothing for output: OpenAI at 10 cents per million input tokens, Jev at 4.2 cents per million.
OpenAI's API mirrors Jev's shape with the same three question types: yes/no predicates, choices and scores. A yes/no query returns a JSON object with a probability field.
Willison had GPT-6 Astra read OpenAI's new API documentation and build the plugin, modeled on his existing llm-typesafe plugin for Jev.

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Read the source simonwillison.net