MTPLX
Hacker ApprovedShares rank 6 of 27 inference engines by recommenders
Inference engine that speeds up local Qwen models on Apple Silicon Macs, often reported 2-3x faster. Some people did not get good results with it.
Inference engine that speeds up local Qwen models on Apple Silicon Macs, often reported 2-3x faster. Some people did not get good results with it.
17 people gave an opinion
- recommend it
- 15
- criticize it
- 2
- approval share
- 88%
Recommended in 5 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: not recommended
- May 2026: not recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 15 people recommend itNeeds 10 or more, met.
- 7.5 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 5 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Speeds up local model inference
“Checkout MTPLX. I use the Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed model and I get around 12 tok/s on a 36GB MBP.”
Comment from 17 Aug 2026 in “Qwen 3.8 27B is excellent, but it defaults to overthinking things” “it really wasn't an exaggeration to say that it ran the model 2-3x faster which was super impressive”
Comment from 19 Jul 2026 in “Qwen3.8 is launching and going open-weight soon” “Try 3.8 27B in MTPLX; I get about 30 tok/s with the same hardware as you.”
Comment from 29 Aug 2026 in “GLM-5.3 is now open-weight” - Comment from 8 Sep 2026 in “Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses”
- Comment from 17 Aug 2026 in “Qwen 3.8 27B is excellent, but it defaults to overthinking things”
Runs Qwen 3.8 models well
“I've been having pretty good luck with the Qwen models with the built in MTP heads”
Comment from 1 Sep 2026 in “Apple caught off guard by AI demand for Mac Mini and Mac Studio” “Try 3.8 27B in MTPLX; I get about 30 tok/s with the same hardware as you.”
Comment from 29 Aug 2026 in “GLM-5.3 is now open-weight” “Qwen 3.8 Flash Next with mtplx on 128GB M5 Max is amazing”
Comment from 5 Oct 2026 in “Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s” - Comment from 8 Sep 2026 in “Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses”
Works on consumer Macs
“Checkout MTPLX. I use the Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed model and I get around 12 tok/s on a 36GB MBP.”
Comment from 17 Aug 2026 in “Qwen 3.8 27B is excellent, but it defaults to overthinking things” “I get around 30 tok/s with it, also on an M1 Max with 64GB.”
Comment from 8 Sep 2026 in “Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses” “mtplx runs it at 25 tok/sec on a M4 Max with 48GB RAM”
Comment from 17 Aug 2026 in “Qwen 3.8 27B is excellent, but it defaults to overthinking things” - Comment from 11 Jun 2026 in “Mtplx 1.0.0”
What people warn about
Each mark is one comment
Did not work well for some
“I didn't get on very well with MTPLX when I tried it a while back”
Comment from 14 Aug 2026 in “Qwen3.8-27B” “I still found the general MLX MTP acceleration to be so poor”
Comment from 16 Sep 2026 in “Show HN: Swift-Qwen3.8-27B, -58.3% thinking, x1.95 speed, accuracy of xhigh”
Other inference engines people recommend
The four most recommended other inference engines, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| vLLM | 112 | 56 | 67% | 12 | — | Approved |
| Strata | 31 | 15 | 67% | 5 | — | Approved |
| DwarfStar 4 | 26 | 2 | 93% | 6 | — | Approved |
| SGLang | 19 | 13 | 59% | 8 | — | — |
| MTPLXThis page | 15 | 2 | 88% | 5 | — | Approved |
What people said
14 of the 19 comments that mention it, newest first. Each links to the original comment.
“Qwen 3.8 Flash Next with mtplx on 128GB M5 Max is amazing”
“like you said, ds4, omlx, mtplx, etc.”
“I still found the general MLX MTP acceleration to be so poor”
“I tried it with the author's 4-bit quant of Qwen 3.8 27B”
“I'll go use that now that I know.”
“I get around 30 tok/s with it, also on an M1 Max with 64GB.”
“MTPLX has a nice interface”
“I've been having pretty good luck with the Qwen models with the built in MTP heads”
“Try 3.8 27B in MTPLX; I get about 30 tok/s with the same hardware as you.”
“mtplx runs it at 25 tok/sec on a M4 Max with 48GB RAM”
“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 didn't get on very well with MTPLX when I tried it a while back”
“it really wasn't an exaggeration to say that it ran the model 2-3x faster which was super impressive”
“replaced oMLX for me on my M5 Pro with 64GB”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.