Community Benchmarks
Download oMLX →Real performance numbers from oMLX users around the world. Submit yours from the app (v0.2.6+).
| # | Chip | RAM | Model | Quant | Ctx | PP tok/s | TG tok/s | Date ↓ |
|---|---|---|---|---|---|---|---|---|
| 154631 | M3 Ultra (80c) | 512 GB | DeepSeek-V4-Flash | 4bit | 1k | 456.8 | 29.0 | 26-04-26 |
| 154632 | M3 Ultra (80c) | 512 GB | DeepSeek-V4-Flash | 4bit | 4k | 62.1 | 26.9 | 26-04-26 |
| 154633 | M3 Ultra (80c) | 512 GB | DeepSeek-V4-Flash | 4bit | 8k | 31.1 | 26.5 | 26-04-26 |
| 154634 | M4 (10c) | 32 GB | Llama-3.2-1B-Instruct | 4bit | 8k | 1,127 | 60.7 | 26-04-26 |
| 154635 | M4 (10c) | 32 GB | Llama-3.2-1B-Instruct | 4bit | 16k | 912.0 | 58.3 | 26-04-26 |
| 154636 | M4 (10c) | 32 GB | Llama-3.2-1B-Instruct | 4bit | 4k | 1,249 | 81.7 | 26-04-26 |
| 154637 | M4 (10c) | 32 GB | Llama-3.2-1B-Instruct | 4bit | 1k | 1,274 | 113.2 | 26-04-26 |
| 154638 | M4 Max (40c) | 128 GB | Qwen3-VL-8B-Instruct | 8bit | 4k | 839.4 | 50.4 | 26-04-26 |
| 154639 | M4 Max (40c) | 128 GB | Qwen3-VL-8B-Instruct | 8bit | 1k | 812.9 | 54.8 | 26-04-26 |
| 154640 | M4 (10c) | 16 GB | Qwen3.5-4B | 4bit | 4k | 273.3 | 29.8 | 26-04-26 |