← Back to community benchmarks
Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated-hermes-oQ4-fp16-mtp
Performance
32k
tokens
1,579
PP tok/s
74.2
TG tok/s
20756
TTFT (ms)
23.3
Peak mem (GB)
Hardware
Chip
M2 Ultra (60c)
Memory
64 GB
GPU Cores
60
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 | 1,968 | 85.7 | 19.9 GB | view |
| 4k | 2,060 | 83.3 | 20.7 GB | view |
| 8k | 1,979 | 81.2 | 21.0 GB | view |
| 16k | 1,835 | 77.8 | 21.8 GB | view |
| 32k | 1,579 | 74.2 | 23.3 GB | current |
| 64k | 1,153 | 63.4 | 25.7 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 85.7 | 1.00× |
| 2× | 143.5 | 1.67× |
| 4× | 365.5 | 4.26× |