youssofal / MTPLX
The fastest way to run Qwen 3.8 Flash Next, Qwen 3.8 27B and Ternary Bonsai 2 27B on a Mac: 125 tok/s in OpenCode on an M5 Max, and a 27B model on 16 GB Macs. Native MTP speculative decoding on Apple Silicon, exact at any temperature. OpenAI and Anthropic compatible local server.
Why this job: topics: apple-silicon, local-llm, llm-inference, local-ai, mlx, inference-engine
What it does
Every number below was measured on a MacBook Pro M5 Max with 128 GB, fans verified at maximum, sampled at the model's own settings. Conditions and sources for every row, with the raw logs where they exist, are at mtplx.com/benchmarks.
Get started
No single install command. Read the steps in the README (opens in new tab)
- Maintenance
- Active
- Last commit 3 days ago
- Stars
- 2,685
- +18 this week
- Licence
- Apache-2.0
- Allows business use
- Downloads for
- Mac
- From the latest release
- Latest version
- v2.12.3
- Released 3 days ago
- Built with
- Python
- 80% of the code
Is it looked after?
- Contributors
- 6
- Last 90 days
- Issues
- 93% closed
- 26 still open
- Pull requests
- 58 merged
- 44 waiting
- Releases
- 57
- 30+ in the last year
Similar repos
| Repo | Stars | Last commit | Licence | Downloads for |
|---|---|---|---|---|
| ARahim3 / mlx-dspark | 736 | yesterday | MIT | No downloads |
| incoai / splash | 2,159 | 3 days ago | Apache-2.0 | Mac |
| carloslfu / slotstream | 659 | today | MIT | No downloads |
More: download files, languages, topics and project record
Languages
- Python 80%
- Swift 18%
- Metal 1%
- TypeScript 1%
Topics
- apple-silicon
- local-ai
- mlx
- mtp
- openai-compatible
- qwen
- speculative-decoding
- anthropic-compatible
- inference-engine
- metal
- qwen3-next
- mtplx
Project record
- JobLocal LLMs
- StatusFeatured
- Commits2,141
- Forks211
- Watching20
- Stars in 28 days+120
- CreatedMay 2, 2026
- First seen by usNov 7, 2026
- CheckedNov 9, 2026