LLM 0.32 RC2 Changes the Default Model and Adds an Endpoint Command
Simon Willison released LLM 0.32rc2 on July 30, changing the default model for users without a configured choice from GPT-4o mini to GPT-5.6 Luna and adding an llm openai endpoint command for arbitrary OpenAI-compatible services. The pre-release also removes a dependency and fixes logging and schema-validation bugs; it builds on RC1's same-day migration to a content-addressed message store, so adopters should back up existing logs before testing.
Simon Willison released LLM 0.32rc2 on July 30 as the second release candidate for version 0.32. The update changes the tool's fallback model, adds a direct path to OpenAI-compatible endpoints, and carries forward a substantial logging redesign introduced in RC1 earlier the same day.
What RC2 changes
Users who have not chosen their own default model will now get GPT-5.6 Luna instead of GPT-4o mini. The release notes list Luna at $0.20 per million input tokens and $1.20 per million output tokens, compared with $0.15 and $0.60 for GPT-4o mini. Users can still set GPT-4o mini, GPT-5 nano, or another available model explicitly.
The new llm openai endpoint command can send prompts, start chats, and list models against an arbitrary OpenAI-compatible endpoint without first adding that service to LLM's model configuration. Those calls are not logged. RC2 also removes the sqlite-migrate dependency because sqlite-utils 4.0 now supplies that functionality, fixes a pending-tool-call reporting bug, and replaces an unhelpful schema-DSL IndexError with a descriptive ValueError for malformed fields.
The logging migration underneath it
RC2 includes the RC1 changes released several hours earlier. That earlier candidate moved new prompt and response records into a content-addressed message store intended to preserve the interaction chain while deduplicating repeated messages. Existing rows in the legacy responses table remain in place, and llm logs reads both generations of data.
The migration also changes what is retained. Raw provider payloads are no longer stored, so provider-only details that lived exclusively in those payloads, including OpenAI completion log probabilities, will not be preserved for new interactions. The release notes recommend backing up the logs database before upgrading.
Why practitioners should test before adopting
The endpoint command lowers setup friction for teams evaluating compatible providers, while the logging schema gives conversations and tool activity a more structured record. The default-model change is also operationally meaningful: scripts that relied on the implicit default can change cost and behavior without changing their command line.
This is still a release candidate, not the stable 0.32 release. The two candidates arrived on the same day, so teams with important log history should test the migration on a backup and pin an explicit default model before rolling it into shared workflows.
Key Points
- 1LLM 0.32rc2 changes the implicit default from GPT-4o mini to GPT-5.6 Luna while preserving explicit model selection.
- 2A new llm openai endpoint command targets arbitrary OpenAI-compatible services without permanent model configuration, and those calls are not logged.
- 3The release carries forward RC1's new content-addressed message store; existing legacy logs remain readable, but new raw provider payloads are no longer retained.
Scoring Rationale
The release candidate materially changes the default model, adds a useful provider-compatibility command, and introduces a consequential logging migration for an established developer tool. Its impact is moderated because 0.32rc2 is pre-release software and the evidence is limited to the maintainer's release records and the authoritative package registry.
Sources
Primary source and supporting public references used for this report.
View 1 more source
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Logistics & Shipping problems
