A community-built plugin, llm-openai-decisions 0.1a0, has been released to provide a Jev-style interface for OpenAI's recently announced Decisions API. The developer who published the plugin already had an llm-typesafe plugin for Jev, and used GPT-6 Astra to read OpenAI's new API documentation and implement llm-openai-decisions inspired by llm-typesafe.
What the OpenAI Decisions API offers and how it compares to Jev
OpenAI's Decisions API — announced at last week's DevDay — is conceptually similar to Jev. Both support three question types: yes/no (predicate), choices, and scores. OpenAI's decision model, gpt-6-luna, additionally supports image input alongside text input, which distinguishes it from Jev in this respect.
Pricing differs as well: OpenAI charges for input tokens at $0.10 (10 cents) per million input tokens, whereas Jev's rate is $0.042 (4.2 cents) per million input tokens. Both systems price input rather than output.
Installing and using the plugin
Install the plugin with:
llm install llm-openai-decisions
An example image-based query from the announcement looks like this:
llm -m openai-decisions/gpt-6-luna \
-a https://static.simonwillison.net/static/2025/two-pelicans.jpg \
-s 'Does this image contain any mammals?'
The example response shown in the announcement is:
{
"type": "predicate",
"name": "evaluation",
"probability": 0.0
}
Refer to the plugin README for complete details on running the other question types.
Why this matters
The plugin provides a quick way for developers familiar with Jev-style interfaces to experiment with OpenAI's Decisions API, including image-capable decision models. The side-by-side comparison of capabilities and input-token pricing helps teams decide which API best fits their needs.
Tags
openai, llm, coding-agents, jev
