← Back to community benchmarks
Qwen3.5-4B
Performance
32k
tokens
1,027
PP tok/s
84.3
TG tok/s
31902
TTFT (ms)
7.3
Peak mem (GB)
Hardware
Chip
M4 Max (40c)
Memory
128 GB
GPU Cores
40
Software
oMLX
v0.3.8
macOS
macOS 15.7.5
Context
32,768
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 1,092 | 132.7 | 3.4 GB | view |
| 4k | 1,373 | 128.8 | 4.1 GB | view |
| 8k | 1,374 | 121.4 | 4.7 GB | view |
| 16k | 1,179 | 103.9 | 5.5 GB | view |
| 32k | 1,027 | 84.3 | 7.3 GB | current |
| 64k | 843.6 | 69.1 | 10.8 GB | view |
| 128k | 511.3 | 41.4 | 18.1 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 132.7 | 1.00× |
| 2× | 188.0 | 1.42× |
| 4× | 280.0 | 2.11× |
| 8× | 241.2 | 1.82× |