community pool / model answer
Qwen3.8-Flash-Next
chat · Qwen/Qwen3.8-Flash-Next
69.4 tok/s · model median · 19 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| Arc Pro B70 32GB ×2 | 3-bit | 21.9 tok/s | reported | - | - GB | 16,384 | llama.cpp |
| Arc Pro B70 32GB ×2 | 4-bit | 24.5 tok/s | reported | 2,810 | - GB | 49,152 | llama.cpp |
| M3 Ultra 512GB | 4-bit | 14 tok/s | reported | 2,053 | - GB | 2,048 | mlx |
| Radeon AI Pro R9700 32GB ×3 | 3-bit | 31.1 tok/s | measured | 1,001 | 64 GB | 131,072 | llama.cpp |
| Radeon AI Pro R9700 32GB ×3 | 4-bit | 7.7 tok/s | reported | 4,017 | - GB | 789 | llama.cpp |
| RTX PRO 6000 Blackwell 96GB ×2 | 8-bit | 125.4 tok/s | measured | 100 | - GB | 262,144 | vllm |
| Ryzen AI Max 395 128GB | 4-bit | 54 tok/s | reported | 6 | 124.1 GB | 262,144 | custom |
Context tested
The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.