← Back to community benchmarks
Qwen3.6-35B-A3B
Performance
4k
tokens
2,636
PP tok/s
89.6
TG tok/s
1554
TTFT (ms)
21.1
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.2.20
macOS
macOS 26.5
Context
4,096
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 1,850 | 91.1 | 20.1 GB | view |
| 4k | 2,636 | 89.6 | 21.1 GB | current |
| 8k | 2,838 | 84.5 | 21.4 GB | view |
| 16k | 2,613 | 73.0 | 22.0 GB | view |
| 32k | 2,219 | 66.4 | 23.4 GB | view |
| 64k | 1,670 | 53.3 | 26.0 GB | view |
| 128k | 1,017 | 35.6 | 31.4 GB | view |
| 195k | 730.2 | 27.5 | 37.1 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 91.1 | 1.00× |
| 2× | 164.0 | 1.80× |
| 4× | 269.0 | 2.95× |
| 8× | 411.8 | 4.52× |