← Back to community benchmarks
Qwen3.5-9B
Performance
32k
tokens
1,259
PP tok/s
35.7
TG tok/s
26021
TTFT (ms)
8.7
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.2.8
macOS
macOS 15.6
Context
32,768
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 1,139 | 52.3 | 6.7 GB | view |
| 4k | 1,324 | 51.5 | 7.0 GB | view |
| 8k | 1,390 | 49.9 | 7.1 GB | view |
| 16k | 1,352 | 51.5 | 7.6 GB | view |
| 32k | 1,259 | 35.7 | 8.7 GB | current |
| 64k | 1,078 | 24.9 | 10.8 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 52.3 | 1.00× |
| 2× | 101.3 | 1.94× |
| 4× | 195.1 | 3.73× |
| 8× | 290.0 | 5.54× |