MTPLX
Fewer than 5 people recommend it so far, so there's no written verdict. Here is what commenters said.
3recommend
1warn against
Jun 11, 2026first mention
Oct 8, 2026latest mention
Why people recommend it
“I’ll go use that now that I know.”
switched from llama-serveron “Benchmarking Qwen3.8 27B quantizations: 4-bit holds …”
“Checkout MTPLX. I use the Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed model and I get around 12 tok/s on a 36GB MBP.”
36GB MacBook Pro useron “Qwen 3.8 27B is excellent, but it defaults to overth…”
Show 1 more
“I have been using MTPLX (replaced oMLX for me)”
Caveats and dissent
“I still found the general MLX MTP acceleration to be so poor”
Often compared with
Who recommends it in their own words
Threads where it came up
- Whistle: Speech to Text in 16.9 MB 1 mention
- Show HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh 1 mention
- Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses 1 mention
- Qwen 3.8 27B is excellent, but it defaults to overthinking things 1 mention
- Mtplx 1.0.0 1 mention
Data: mtplx.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 5
“local Qwen 3.8 running under MTPLX”
“I still found the general MLX MTP acceleration to be so poor”
“I’ll go use that now that I know.”
“Checkout MTPLX. I use the Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed model and I get around 12 tok/s on a 36GB MBP.”
“I have been using MTPLX (replaced oMLX for me)”