LLM 0.32rc2 Release: GPT-5.6 Luna as New Default, OpenAI Endpoint Command Added
The latest release, LLM 0.32rc2, introduces significant updates including a dependency fix and two key new features. The default model for users who haven't specified their own is now GPT-5.6 Luna, replacing GPT-4o mini. Luna offers improved performance but comes with higher costs: $0.20 per million input tokens and $1.20 per million output tokens, compared to $0.15 and $0.60 for GPT-4o mini. Users can revert to GPT-4o mini or opt for the even more economical GPT-5 nano ($0.05/$0.40) via command line. A notable addition is the `llm openai endpoint` command, which allows users to run prompts, chats, and list models against any OpenAI-compatible endpoint without prior configuration. These interactions are not logged. This new command addresses the need for a straightforward CLI tool to test prompts against local or third-party OpenAI-compatible services. An example is provided for running a prompt against a local LM Studio model using a `uvx` one-liner, querying LLM version and time.
This release highlights the ongoing evolution of LLM interfaces and the competitive landscape of model pricing and performance. The shift to GPT-5.6 Luna as the default suggests a move towards more capable, albeit more expensive, models, reflecting a market trend where advanced features command a premium. The introduction of the `llm openai endpoint` command democratizes access to various LLM backends, reducing friction for developers and researchers experimenting with different models and hosting solutions. This feature fosters interoperability and innovation by allowing seamless interaction with diverse OpenAI-compatible APIs, potentially accelerating the adoption of specialized or locally hosted models. The explicit mention of non-logged calls addresses privacy concerns, a critical factor in user adoption and trust.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.