You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
mac: default to qwen3-vl:30b-a3b-instruct (vision + coding + tools)
Switch the Mac app and bash launcher default model to qwen3-vl:30b-a3b-instruct
after deleting the qwen2.5-coder 7b/14b models locally. It's a ~3B-active MoE, so
prefill stays fast, and it adds vision (the shim already forwards image blocks)
and reliable tool calls — verified coding, tool_use, and image description all
work through the shim. Drops the haiku->7b split in the launcher (only one model
now) and reseeds the Edit-Model-Map template accordingly. Windows app/launcher
stay on qwen2.5-coder — that box keeps its smaller models for its RAM/sim setup.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
0 commit comments