The release of LLM version 0.32 introduces significant enhancements, notably visible reasoning traces, server-side tools, redesigned SQLite logs, and broader functionality with the OpenAI Responses API. This update marks a pivotal advancement since the project's inception, optimizing user interactions with language models.
Key improvements for LLM Command Line Interface (CLI) users include the ability to display reasoning traces for reasoning models directly to standard error. Users can now observe the internal thinking processes of these models without affecting standard output, with the option to suppress this feature using the -R/--hide-reasoning flag. The LLM now supports the GPT-5.6 model family, with the cost-effective GPT-5.6 Luna designated as the new default model for prompts.
Additional features allow LLM to integrate with server-side tools from multiple providers. Notably, OpenAI offers a code execution environment, enabling commands like `llm --tool CodeInterpreter 'Show current python and SQLite versions'`. Furthermore, the llm-anthropic plugin has been expanded to include tools such as WebSearch, WebFetch, CodeExecution, and AnthropicMCP, allowing for streamlined interactions with various APIs in a single request-response format.
The Python API has also seen enhancements, transitioning from a model that required individual message handling to one that accommodates a complete history of messages in each request. This change reflects a more accurate representation of how LLMs function, facilitating a more efficient user experience.
Overall, LLM 0.32 introduces key technical improvements that enhance both CLI and API user experience, expanding the capabilities of developers who rely on these tools for specialized applications.