CLI Tool LLM 0.37 Cuts Default Costs in Half

According to simonwillison.net, developer Simon Willison released version 0.37 of his popular CLI tool and Python library for interacting with large language models. This update introduces two months of accumulated fixes and developer-focused features, most notably dropping default operational costs.

What changes in your terminal workflow?

The release changes the default fallback model to gpt-6-luna, which sits at exactly half the price of the previous default, gpt-5.6-luna. For developers running frequent queries or automated scripts from the command line, this directly slashes token overhead without requiring manual configuration changes.

Additionally, simonwillison.net highlights a new command: llm openai endpoint add. This utility simplifies the process of configuring extra models that rely on OpenAI-compatible APIs, removing the friction of manual config file editing when switching providers.

How does multimodal embedding work now?

The tool's embedding support now accepts distinct content types for binary inputs. According to simonwillison.net, this adjustment is essential for models like Gemini Embedding 2, which process text, audio, images, and video within embedding pipelines.

How to install or update

To upgrade your environment to the latest version, run your standard Python package manager update command for the tool:

• Open your terminal
• Run pip install --upgrade llm or your preferred package manager equivalent
• Verify the installation by running llm --version

Is it worth adopting for your stack?

If you build workflows around terminal-based AI calls, version 0.37 is a straightforward upgrade. Interestingly, Willison noted that nearly all contributions in this release came from AI-assisted pull requests, framing them not as a burden, but as detailed bug reports with attached code examples.