Good to see this shipped as open weights. Cache-aware routing is the part that matters most. Coding agent prefixes grow monotonically, so every model switch pays full uncached input again. Optimizing on list price alone can cost more than never routing.

→ View original post on X — @akshay_pachaar