← Back to community benchmarks
Qwen3.6-27B
Performance
8k
tokens
429.9
PP tok/s
12.2
TG tok/s
19056
TTFT (ms)
53.8
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.3.7
macOS
macOS 26.4.1
Context
8,192
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 374.1 | 12.6 | 51.7 GB | view |
| 4k | 431.1 | 12.4 | 53.2 GB | view |
| 8k | 429.9 | 12.2 | 53.8 GB | current |
| 16k | 421.0 | 11.9 | 55.1 GB | view |
| 32k | 395.2 | 11.2 | 57.6 GB | view |
| 64k | 342.5 | 0.3 | 67.8 GB | view |
| 128k | 258.4 | 8.1 | 73.1 GB | view |
| 195k | 205.1 | 0.1 | 111.3 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 12.6 | 1.00× |
| 2× | 13.7 | 1.09× |
| 4× | 27.4 | 2.17× |
| 8× | 52.4 | 4.16× |