← Back to community benchmarks
Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved
Performance
16k
tokens
1,123
PP tok/s
37.4
TG tok/s
14590
TTFT (ms)
67.6
Peak mem (GB)
Hardware
Chip
M3 Max (40c)
Memory
128 GB
GPU Cores
40
Software
oMLX
v0.4.1
macOS
macOS 26.0.1
Context
16,384
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 880.3 | 33.4 | 65.6 GB | view |
| 4k | 1,320 | 34.6 | 66.4 GB | view |
| 8k | 1,279 | 42.3 | 66.7 GB | view |
| 16k | 1,123 | 37.4 | 67.6 GB | current |
| 32k | 887.9 | 28.7 | 69.1 GB | view |
| 64k | 154.2 | 9.3 | 72.0 GB | view |
| 128k | 124.8 | 9.5 | 78.1 GB | view |
| 195k | 112.3 | 7.1 | 85.0 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 33.4 | 1.00× |
| 2× | 17.1 | 0.51× |
| 4× | 27.2 | 0.81× |
| 8× | 30.9 | 0.93× |