← Back to community benchmarks
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinkingmxfp8
Performance
4k
tokens
275.5
PP tok/s
14.4
TG tok/s
14865
TTFT (ms)
41.8
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.3.4
macOS
macOS 26.3.1
Context
4,096
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 258.4 | 14.8 | 39.8 GB | view |
| 4k | 275.5 | 14.4 | 41.8 GB | current |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 14.8 | 1.00× |
| 2× | 25.7 | 1.74× |
| 4× | 28.7 | 1.94× |
| 8× | 31.2 | 2.11× |