← Back to community benchmarks
Qwen3.6-35B-A3B-PARO-packed
Performance
16k
tokens
3,195
PP tok/s
73.1
TG tok/s
5128
TTFT (ms)
22.3
Peak mem (GB)
Hardware
Chip
M5 Max (40c)
Memory
128 GB
GPU Cores
40
Software
oMLX
v0.4.2rc1
macOS
macOS 26.6
Context
16,384
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 2,162 | 78.1 | 20.5 GB | view |
| 4k | 3,410 | 70.5 | 21.3 GB | view |
| 8k | 3,240 | 80.6 | 21.6 GB | view |
| 16k | 3,195 | 73.1 | 22.3 GB | current |
| 32k | 2,704 | 68.6 | 23.6 GB | view |
| 64k | 1,983 | 59.4 | 26.2 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 78.1 | 1.00× |
| 2× | 115.9 | 1.48× |
| 4× | 232.6 | 2.98× |