MTPLX: the fastest way to run Qwen 3.8 on a Mac. Native multi-token-prediction speculative decoding on Apple Silicon, two to three times the speed of plain decoding, exact at any temperature.
Qwen 3.8 27B Optimized Speed
4-bit dynamic quant. Great coding speeds and good quality. Recommended.
This is the model MTPLX recommends for coding on a modern Mac with 32 GB or more. It is Qwen3.8-27B with its native multi-token-prediction head kept, so MTPLX can draft several tokens ahead and verify them in one pass. Same model, same output distribution, a lot more tokens per second.
Measured on MTPLX 2.11.3 (16 September 2026)
This is a Qwen3.8-27B MLX pack for MTPLX, the fastest way to run Qwen 3.8 27B on a Mac. All numbers on a MacBook Pro M5 Max, fans verified at maximum, sampled at the model's own settings. Conditions and sources: mtplx.com/benchmarks.
| Run | tok/s |
|---|---|
| Optimized Speed rewriting a file it just wrote, stock settings (MTPLX 2.10.0) | 87.6 |
| Bare Speed on a fresh coding task at official Qwen 3.8 sampling (MTPLX 2.7.0) | 65.2 |
| Optimized Speed on a 3k-token chat answer (MTPLX 2.10.0) | 64.3 |
| The 27B record on a fresh generation: Qwen 3.6 27B Optimized Speed, 192-token bench, 2 July 2026, raw logs published | 81.74 |
Pack quality against the bf16 Qwen3.8-27B checkpoint, teacher-forced over 2,389 positions of code, prose, JSON and a multilingual notice (16 September 2026): Optimized Speed (4-bit dynamic) 96.0 percent top-1 agreement and KL 0.012; Optimized Quality (8-bit dynamic) 99.3 percent and KL 0.0005. Exactness on this release: a thousand four-token draws from the fast path match a thousand from the plain path within the plain path's own noise, at temperature 1, top-p 0.95, top-k 20. Details: MTPLX 2.11.3 release notes.
Runs on Apple Silicon Macs with 32 GB of unified memory or more (36 GB for Optimized Quality): MacBook Pro, MacBook Air, Mac mini and Mac Studio on M1 to M5. Guide: Run Qwen 3.8 27B on a Mac. Comparison pages: MTPLX vs mlx-serve, MTPLX vs oMLX, MTPLX vs LM Studio, MTPLX vs Ollama.
Speeds
Measured on an M5 Max, fans verified at max, single stream, generation running to the model's own stop, official Qwen 3.8 sampling (temperature 1.0, top-p 0.95, top-k 20).
| Run | tok/s |
|---|---|
Coding task, medium reasoning, mtplx serve | 58.7 |
| Same task inside the MTPLX Mac app | 55.5 |
| Long reasoning at xhigh, 28k and 20k token answers | 35.1 and 37.3 |
For context on the same night and the same task: the previous MTPLX flagship, Qwen 3.6 27B Optimized Speed V2, ran 59.9 to 60.1 tok/s, and oMLX 0.5.7 with its own Qwen 3.8 4-bit MTP quant ran 63.3 tok/s. MTPLX Bare Speed ran 65.2 on that task; this build gives up a few tok/s to be much closer to the original model.
Draft acceptance on the coding task by depth: 0.95, 0.88, 0.80. Verify cost 50 to 53 ms per round.
How it is built
A dynamic 4-bit quant, the same hand-tuned layout as our Qwen 3.6 Optimized Speed V2:
- The bulk of the model is 4-bit with 32-weight groups.
- The parts that hurt most at 4-bit are kept at 8-bit: the embeddings, the output head, all 48 GDN output projections, and the last 8 MLP blocks.
- Sensitive parts stay 16-bit: the GDN convolution kernels and recurrent state parameters, every norm, and the whole MTP head.
- KL divergence to the original bf16 model on our coding battery: 0.0220. That is 1.7x closer to the original than the flat 4-bit Bare Speed build.
| Download | 20.4 GB |
| Peak unified memory (measured, this artifact) | 23.6 GB |
| Context window | 262,144 tokens |
| MTP depth | 3 |
| Sampling | temperature 1.0, top-p 0.95, top-k 20 (the official Qwen 3.8 contract) |
The tuned depth and draft settings ship inside mtplx_runtime.json. MTPLX
reads them on load. No flags needed. Reasoning effort levels (xhigh, medium,
low) work, and preserved thinking flows through the MTP path like any other
tokens.
Speculation in MTPLX is exact. Drafts are accepted with the probability-ratio rule plus residual resampling, so what you sample is what the model would have sampled without speculation, at any temperature.
Use it
Mac app: download at mtplx.com, pick "Qwen 3.8 27B Optimized Speed". It is the default on modern Macs.
Command line:
pip install mtplx
mtplx serve --model Youssofal/Qwen3.8-27B-MTPLX-Optimized-Speed
Then point OpenCode, Pi, Claude Code, Cline, or anything that speaks the OpenAI
or Anthropic API at http://127.0.0.1:8000.
Siblings: Bare Speed (quickest burst chat speeds, lower quality) and Optimized Quality (8-bit, perfect quality). On an M1 or M2 Mac use the FP16 build of this model.
On a Mac with 16 GB, or to use half the memory, try Bonsai 2 27B Optimized Speed, Prism ML's ternary Qwen 3.8 27B: 64.4 tok/s after a 4,061-token prompt against 52.6 for this pack, with a peak of 11.4 GB against 23.9 (MTPLX 2.12.0, M5 Max).