llm 0.33
Blog · Watch/Read · (gathered 2026-08-29)
llm 0.33introduces several enhancements to thellmcommand-line tool and Python library, including dependency upgrades and new user-facing features.- The
llm embedandllm embed-multicommands, along with their Python API methods, now accept a--keyargument for specifying per-call keys for embedding plugins, aligning with how regular LLM models handle keys. - Users can now repeat the
llm prompt -t/--templateoption to combine multiple templates, enabling a pattern where model configurations and options are packaged in one template and combined with prompts from another. - Reasoning-capable Responses API models now support a
reasoning_summaryoption (withauto,concise, ordetailedvalues) viallm openai endpoint --responses, which is particularly useful for evaluating different models that emulate the OpenAI Responses API.