New release of LLM adds support for reasoning traces, OpenAI Responses, server-side tools, and smarter logging
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
LLM 0.32 now streams structured events (reasoning, text, tool calls, images) instead of just strings, breaking any code that assumed plain-text output. This forces you to rewrite streaming loops to handle multi-modal responses, but lets you log or filter reasoning traces without polluting downstream pipelines.
LLM 0.32 adds real-time reasoning traces to standard error, letting engineers debug agent logic without contaminating output streams—critical for piping to downstream tools. This enables inspection of model "thinking" during chained operations while maintaining clean production outputs, directly addressing a major pain point in debugging complex agent workflows. Server-side tool integrations (like OpenAI's CodeInterpreter) now work natively, allowing safer execution of sandboxed code and web queries within a single API call.