← Back to community benchmarks
Qwen3___5-27B
Performance
4k
tokens
421.3
PP tok/s
21.0
TG tok/s
9722
TTFT (ms)
29.5
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.2.6
macOS
macOS 26.3.1
Context
4,096
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 413.1 | 21.6 | 29.1 GB | view |
| 4k | 421.3 | 21.0 | 29.5 GB | current |
| 8k | 417.3 | 20.4 | 29.9 GB | view |
| 16k | 404.4 | 18.9 | 30.8 GB | view |
| 32k | 379.2 | 17.0 | 32.6 GB | view |
| 64k | 334.1 | 14.5 | 36.2 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 21.6 | 1.00× |
| 2× | 38.4 | 1.78× |
| 4× | 57.5 | 2.66× |