v0.38.5

Mesh-LLM/mesh-llmv0.38.5Mar 23, 2026by michaelneale

AI Summary

Fixed 400 errors occurring during long conversations by increasing the small-host context window. Router logic was updated to stop preferring small models and tighten the random spread window.

Key Highlights

  • Fixed chat 400 errors on long conversations
  • Increased small-host context from 8K to 16K
  • Router stops preferring small models for chat
  • Tightened random model spread window (50→15 points)

New Features

  • Context window expansion
  • Router logic update
  • New Qwen3.5-9B router profile

Full Release Notes

- Fix chat 400 errors on long conversations: bump small-host context from 8K to 16K
- Router: stop preferring small models for chat — higher tier always preferred
- Tighten random model spread window (50→15 points) so weak models are fallbacks, not equal contenders
- Add Qwen3.5-9B router profile (tier 2)
- TODO: context-aware routing, retry-on-400 noted as future work