https://github.com/jundot/omlx/releases Go to model down Multiple options, depending on your RAM 35B parameters with 3 billion active:
unsloth/Qwen3.6-35B-A3B-UD-MLX-3bit
- 17.4 GB (36GB+ RAM ideal)unsloth/Qwen3.6-35B-A3B-UD-MLX-4bit
- 21.6 GB (48GB+ RAM ideal)unsloth/Qwen3.6-35B-A3B-MLX-8bit
- 37.7GB GB (64GB+ RAM ideal)
27B billion parameters
unsloth/Qwen3.6-27B-UD-MLX-4bit
- 26.2GB (48GB+ RAM ideal)unsloth/Qwen3.6-27B-UD-MLX-6bit
- 30.5GB (64GB+ RAM ideal)unsloth/Qwen3.6-27B-UD-MLX-8bit
- 34.7GB (64GB+ RAM ideal)
- Go to model settings
- Pin and default model to the downloaded one
- Open the model's settings
- Enable TurboQuant KV Cache in3.5-bit
- Go to global settings
- Turn on Fallback to Default Model
- Set Hot Cache Limit (In-Memory Cache) to 10% - Set Cold Cache Limit (SSD Cache) to 10% - Increase Max Context Window to256000
- Increase Max Tokens to64000
- Save
Add "CLAUDE_CODE_ATTRIBUTION_HEADER": "0" inenv key inside~/.claude/settings.json
(Ref)Example:
{ "env": { "CLAUDE_CODE_ATTRIBUTION_HEADER": "0" } }
source & further reading
gist.github.com — original article
agent-proxy — a zero-dep proxy that shows the bloat in Claude Code's requests (ranked tool table + full readable Markdown of every request)
PowerShell profile and test suite
A plan-of-plans for a production-grade end-to-end e-commerce website.