← Back to community benchmarks
Qwen3.5-4B
Performance
16k
tokens
1,355
PP tok/s
98.7
TG tok/s
12090
TTFT (ms)
5.3
Peak mem (GB)
Hardware
Chip
M2 Ultra (76c)
Memory
128 GB
GPU Cores
76
Software
oMLX
v0.2.6
macOS
macOS 26.3
Context
16,384
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 1,428 | 137.5 | 4.0 GB | view |
| 4k | 1,440 | 128.2 | 4.2 GB | view |
| 8k | 1,414 | 117.4 | 4.6 GB | view |
| 16k | 1,355 | 98.7 | 5.3 GB | current |
| 32k | 1,247 | 74.5 | 6.9 GB | view |
| 64k | 1,051 | 50.4 | 8.6 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 137.5 | 1.00× |
| 2× | 206.8 | 1.50× |
| 4× | 255.6 | 1.86× |
| 8× | 301.3 | 2.19× |