Default Models

This page documents gmuse’s default LLM model choices and the rationale.

Default mapping (economy-focused)

• openai: gpt-4o-mini

• anthropic: claude-haiku-4-5

• cohere: command-light

• azure: gpt-4o-mini

• gemini: gemini/gemini-flash-lite-latest

Rationale

The defaults prefer smaller/light/mini/haiku variants intended to be lower-cost and lower-latency while still providing reliable instruction-following for short text generation tasks such as commit message generation. Users can override these defaults using the GMUSE_MODEL environment variable or the --model CLI option for a single invocation.

Notes

• These defaults may be updated over time as providers introduce newer, more cost-efficient variants. If you need higher quality or advanced capabilities (e.g., long-form reasoning or large-context agents), specify a stronger model via --model or config file.