Qwen3.5 35B
15 records across 4 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX 5090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 165.2 | 6605.2 | – | – | 4096 | – | sourced | – | source |
| RTX 5090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 148.3 | 6142.3 | – | – | 16384 | – | sourced | – | source |
| RTX 5090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 143.2 | 5611.4 | – | – | 32768 | – | sourced | – | source |
| RTX 5090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 133.5 | 4624.5 | – | – | 65536 | – | sourced | – | source |
| RTX 5090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 118.2 | 3242.6 | – | – | 131072 | – | sourced | – | source |
| RTX 3090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 111.2 | 2622.1 | – | – | 4096 | – | sourced | – | source |
| RTX 3090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 107.1 | 2381.3 | – | – | 16384 | – | sourced | – | source |
| RTX 3090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 101.2 | 2121.6 | – | – | 32768 | – | sourced | – | source |
| RTX 5090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 97.3 | 2003.7 | – | – | 262144 | – | sourced | – | source |
| RTX 3090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 93.1 | 1749.8 | – | – | 65536 | – | sourced | – | source |
| RTX 3090 | Qwen3.5 35B 35B | MXFP4 | llama.cpp | 79.4 | 1288.9 | – | – | 131072 | – | sourced | – | source |
| DGX Spark | Qwen3.5 35B | unspecified | llama.cpp (CUDA) | 63.5 | – | – | – | tg128 | 2026-04-01 | community | – | source |
| DGX Spark | Qwen3.5 35B | unspecified | llama.cpp (CUDA) | 50.1 | – | – | – | tg128 | 2026-04-01 | community | – | source |
| 3x RTX PRO 4000 Blackwell | Qwen3.5 35B | unspecified | CUDA | – | 3701 | – | – | 65536 | 2026-10 | community | – | source |
| 3x RTX PRO 4000 Blackwell | Qwen3.5 35B | unspecified | CUDA | – | 7148 | – | – | 65536 | 2026-10 | community | – | source |