Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
Data Science
Simon Willison's Blog·Jul 30·5 min read
llm 0.32rc2
TL;DR
The latest release of LLM (0.32rc2) introduces GPT-5.6 Luna as the new default model and a command for using arbitrary OpenAI-compatible endpoints. This enhances user experience with improved model performance and flexibility.
✦ Why It Matters
Switch to GPT-5.6 Luna today for improved model performance in your applications.
Key Takeaways
- Switch to GPT-5.6 Luna for better performance if cost allows.
- Use llm openai endpoint for quick prompt testing without setup.
- Revert to GPT-4o mini or switch to GPT-5 nano for cost savings.
- Experiment with local models using the provided command line example.
Full Summary
LLM 0.32rc2 updates the default model for users who haven't set their own to GPT-5.6 Luna, which offers improved performance over the previous GPT-4o mini, albeit at a higher cost of $0.20 per million input tokens and $1.20 for output tokens. Users can revert to the cheaper GPT-4o mini or switch to the even more economical GPT-5 nano model at $0.05 and $0.40 respectively.
A new command, llm openai endpoint, allows users to run prompts against any OpenAI-compatible endpoint without needing to configure a model first. This feature is particularly useful for testing and experimentation, as it can be executed without installing LLM.
A simple command line example demonstrates how to query a local model using this new functionality, making it easier for developers to interact with language models.