← Back to community benchmarks
Qwen3.6-35B-A3B-oQ6e-fp16-mtp
User commentv0.5.1 5a39ba3a without custom kernels
Performance
32k
tokens
1,898
PP tok/s
74.0
TG tok/s
17263
TTFT (ms)
31.0
Peak mem (GB)
Hardware
Chip
M2 Ultra (76c)
Memory
192 GB
GPU Cores
76
Software
oMLX
v0.5.1
macOS
macOS 26.5.2
Context
32,768
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 2,359 | 83.4 | 27.8 GB | view |
| 4k | 2,505 | 82.9 | 28.6 GB | view |
| 8k | 2,391 | 81.5 | 29.0 GB | view |
| 16k | 2,173 | 78.6 | 29.7 GB | view |
| 32k | 1,898 | 74.0 | 31.0 GB | current |
| 64k | 1,340 | 66.2 | 34.4 GB | view |
| 128k | 757.8 | 55.9 | 39.8 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 83.4 | 1.00× |
| 2× | 139.3 | 1.67× |
| 4× | 203.0 | 2.43× |