llm 0.32rc2 Release: New Default Model and OpenAI Endpoint Command

The latest release candidate for the llm command-line tool, version 0.32rc2, changes its default model and adds a strong new feature for developers. It upgrades the default from GPT-4o mini to the newer GPT-5.6 Luna, which has a slightly higher cost. This release also adds the ‘llm openai endpoint’ command. It lets you work directly with any OpenAI-compatible API without any setup.

Key takeaways

• The default model in llm has changed from GPT-4o mini to GPT-5.6 Luna.
• GPT-5.6 Luna costs more than the previous default but is called a much better model.
• A new ‘llm openai endpoint’ command allows testing prompts against arbitrary OpenAI-compatible APIs without setup.
• Calls made with the new endpoint command are not logged by the LLM tool.
• The release also fixes a dependency issue from the previous release candidate.

New Default Model: GPT-5.6 Luna

The llm 0.32rc2 release changes the default model for users who have not set their own preference from GPT-4o mini to GPT-5.6 Luna. The release notes call Luna a “much better and more recent model” than the old default.

Pricing and Alternatives

This upgrade comes with a slight cost increase. GPT-5.6 Luna is priced at $0.20 per million input tokens and $1.20 per million output tokens. For comparison, GPT-4o mini costs $0.15 and $0.60 for the same input and output tokens, respectively.

Users who wish to switch back to the previous model can do so by running the command llm models default gpt-4o-mini. For an even more economical option, the command llm models default gpt-5-nano sets GPT-5 nano as the default, which costs $0.05 per million input tokens and $0.40 per million output tokens.

Introduction of the ‘llm openai endpoint’ Command

A key new feature in llm 0.32rc2 is the llm openai endpoint command. It allows users to run prompts, chats, and model listings against arbitrary OpenAI-compatible endpoints without first configuring a model. These specific API calls are not logged by the tool.

The feature was added to fill a gap in CLI tools for testing prompts against imitation OpenAI Chat Completions endpoints. You can use the command without installing LLM via a uvx one-liner. The provided example shows using it with tools like llm_version and llm_time to run a prompt against a local LM Studio model.

Fixes and Dependency Resolution

This release of llm 0.32rc2 follows directly on from RC1. The update is a fast iteration, with its new features and fixes building directly on the earlier release candidate.

The main change in this version is a fix for a dependency issue that was present in the previous release candidate. This fix is the core maintenance update for the iteration.

Developer Motivation and Use Cases

Simon Willison added the new ‘llm openai endpoint’ command to the LLM tool out of frustration with the lack of an obvious command-line interface tool for experimenting with arbitrary OpenAI-compatible endpoints. He integrated this capability directly into LLM for developer convenience, and these specific calls are not logged by the system.

A main use case shows its practical application for developers working with diverse LLM endpoints. The feature allows running a prompt with tools against a local model hosted in LM Studio using a single command. The provided example is a one-liner that uses uvx to query a local Gemma model for the current LLM version and time, showing immediate utility without requiring prior model configuration within LLM.