← Back to community benchmarks
Qwen3___6-35B-A3B
Performance
8k
tokens
2,646
PP tok/s
76.1
TG tok/s
3096
TTFT (ms)
36.5
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.3.6
macOS
macOS 26.4.1
Context
8,192
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 1,615 | 81.6 | 35.4 GB | view |
| 4k | 2,523 | 79.8 | 36.2 GB | view |
| 8k | 2,646 | 76.1 | 36.5 GB | current |
| 16k | 2,526 | 71.7 | 37.1 GB | view |
| 32k | 2,187 | 63.0 | 38.5 GB | view |
| 64k | 1,652 | 50.4 | 41.2 GB | view |
| 128k | 1,032 | 36.8 | 46.5 GB | view |
| 195k | 733.8 | 29.3 | 52.2 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 81.6 | 1.00× |
| 2× | 138.5 | 1.70× |
| 4× | 209.2 | 2.56× |
| 8× | 323.2 | 3.96× |