LLM 0.32 Adds Server-Side Tool Support

Simon Willison's open-source LLM CLI has been updated to version 0.32. This release includes reasoning trace display for reasoning models, server-side tools, a revamped log storage structure, and OpenAI Responses API features.

Key changes are as follows:

• Reasoning traces for reasoning models are displayed on standard error (stderr) and can be hidden with the -R/--hide-reasoning option.
• The GPT-5.6 model family is supported by default, with the default model changed to the cheaper GPT-5.6 Luna.
• Server-side tools such as OpenAI's Code Interpreter and WebSearch can be used from the CLI.
• Support for Anthropic's WebSearch, WebFetch, CodeExecution, and AnthropicMCP has been added to the llm-anthropic plugin.
• The llm openai endpoint command allows sending prompts to any OpenAI-compatible endpoint without additional configuration.
• A model.prompt(messages=[]) interface has been added to the Python API to pass a full message array directly.
• The content-addressed SQLite log structure has been redesigned to improve log management features.

With these updates, LLM CLI can now execute agentic tasks, including provider-side tools such as web search, code execution, and MCP integration, alongside model calls, all within a single interface.