Simon Willison has released version 0.32rc2 of his open source LLM command line tool, fixing a dependency problem and updating the default model for new users. The tool now defaults to OpenAI‘s GPT-5.6 Luna, replacing the previous GPT-4o mini selection. This change increases costs to $0.20 per million input tokens and $1.20 per million output tokens, though users can revert to the cheaper GPT-4o mini or the GPT-5 nano model via specific commands.
The most significant addition is a new command that allows immediate testing against arbitrary OpenAI compatible endpoints without prior configuration. This feature addresses a gap in local development workflows by enabling direct interaction with models running on private servers or local machines like LM Studio. The command supports tool usage and logging suppression, making it suitable for rapid experimentation without installing the full application.
* Default model cost increased to $0.20 per million input tokens
* New command supports unconfigured OpenAI compatible endpoints
* Users can switch defaults using `llm models default`
