DeepSeek-V4-Flash-0731-MXFP4
Deepseek family · 304.0B params
1recorded run
26.4best tok/s
1verified
Fastest GPUs for this model (best tok/s)
- RTX 509026.4
All benchmarks Methodology
| GPU | Model | Quant | Backend | Ctx |
tok/s | prompt tok/s | VRAM GB | Source |
| RTX 5090 · 32 GB | DeepSeek-V4-Flash-0731-MXFP4 | MXFP4 | llama.cpp 4821 (c46ffaa5) | 196608 | 26.4 ✓ | 174.5 | 28.3 | src |