community pool / model answer
Step-3.7-Flash-GGUF
chat · stepfun-ai/Step-3.7-Flash-GGUF
30.7 tok/s · model median · 9 runs in the model record.
| Reference rig | Quantization | Median result | Basis | TTFT (ms) | Peak VRAM | Max context | Engines |
|---|---|---|---|---|---|---|---|
| Arc Pro B70 32GB ×4 | 4-bit | 20.8 tok/s | reported | - | 95.6 GB | 32,768 | llama.cpp |
| GB10 Grace Blackwell 128GB | 4-bit | 23 tok/s | reported | 8,470 | - GB | 2,048 | llama.cpp |
| RTX 3090 24GB ×8 | 4-bit | 52.7 tok/s | reported | 199 | - GB | 262,144 | llama.cpp |
| Ryzen AI Max 395 128GB | 4-bit | 22 tok/s | reported | 36,084 | - GB | 65,536 | llama.cpp |
| Ryzen AI Max 395 64GB | 1-bit | 30.8 tok/s | measured | 3,334 | 53.8 GB | 4,096 | llama.cpp |
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.