The LLM toolkit has been updated to version 0.33, bringing several enhancements focused on embedding models and prompt templating.

The most significant change involves upgrading the underlying OpenAI Python library to version 3.x and switching from httpx to httpx2 for HTTP requests—a comprehensive fix addressing issues introduced in a previous release. This update ensures compatibility with newer OpenAI features while improving performance.

For embedding models, users can now specify a --key parameter, allowing for per-call key management without affecting shared model state. This feature improves plugin integration and provides greater control over API access. Existing plugins that rely on self.key will continue to function through a compatibility fallback.

The new release also introduces the ability to combine prompt templates using the -t/--template flag, enabling users to create modular configurations that package models with specific options. For example:

llm -m gpt-5.6-luna -o reasoning_effort high --save lhigh llm "Generate an SVG of a pelican riding a bicycle" --save pelican
# Combine and run the templates
llm -t lhigh -t pelican

Finally, reasoning-capable models now support --responses with options for auto, concise, or detailed summaries—particularly useful when evaluating different model implementations of the OpenAI Responses API.