v0.13.3 v0.13.2

v0.13.3

Sep 30, 2026· 21 days ago

This release extends the output-limit fix to all providers, ensuring that models with smaller output caps (such as OpenRouter, Together, and Fireworks) are recognized by the router and kept in rotation rather than marked as unable to use tools.

v0.13.2

Sep 29, 2026· 23 days ago

Fixes Claude Code compatibility by teaching the router to learn model output limits from refusal messages, allowing models that stop at lower token counts (like Groq's gpt-oss-20b) to continue working correctly.