LLM 0.32 adds reasoning traces, OpenAI Responses support, server-side tools, and smarter logging

LLM 0.32 was released on 4th August 2026, adding significant features. It now displays reasoning traces to standard error when running reasoning models, with an option to hide them. It includes out-of-the-box support for the GPT-5.6 model family, with GPT-5.6 Luna as the new default model. LLM can use server-side tools like OpenAI's CodeInterpreter and WebSearch, and the llm-anthropic plugin adds WebSearch, WebFetch, CodeExecution, and AnthropicMCP tools. The new 'llm openai endpoint' command allows one-liner prompts against any OpenAI-compatible endpoint, which are not logged. The Python API now supports a model.prompt(messages=[]) parameter and a stream_events() method for handling reasoning, text, tool calls, and image attachments. The new llm-chat-completions-server plugin provides a chat completions API. Logging was redesigned using a content-addressable message store modeled after Git. Existing plugins continue to work but need upgrades for full streaming events support. The author notes that LLM is becoming an agent framework, with features like human approval pauses and resumed message history.