Qwen3.5 35B

15 records across 4 hardware.

hardwaremodelquantbackendtok/spp tok/sWttft sctx/tgdateclassflagssource
RTX 5090Qwen3.5 35B 35BMXFP4llama.cpp165.26605.2––4096–sourced–source
RTX 5090Qwen3.5 35B 35BMXFP4llama.cpp148.36142.3––16384–sourced–source
RTX 5090Qwen3.5 35B 35BMXFP4llama.cpp143.25611.4––32768–sourced–source
RTX 5090Qwen3.5 35B 35BMXFP4llama.cpp133.54624.5––65536–sourced–source
RTX 5090Qwen3.5 35B 35BMXFP4llama.cpp118.23242.6––131072–sourced–source
RTX 3090Qwen3.5 35B 35BMXFP4llama.cpp111.22622.1––4096–sourced–source
RTX 3090Qwen3.5 35B 35BMXFP4llama.cpp107.12381.3––16384–sourced–source
RTX 3090Qwen3.5 35B 35BMXFP4llama.cpp101.22121.6––32768–sourced–source
RTX 5090Qwen3.5 35B 35BMXFP4llama.cpp97.32003.7––262144–sourced–source
RTX 3090Qwen3.5 35B 35BMXFP4llama.cpp93.11749.8––65536–sourced–source
RTX 3090Qwen3.5 35B 35BMXFP4llama.cpp79.41288.9––131072–sourced–source
DGX SparkQwen3.5 35B unspecifiedllama.cpp (CUDA)63.5–––tg1282026-04-01community–source
DGX SparkQwen3.5 35B unspecifiedllama.cpp (CUDA)50.1–––tg1282026-04-01community–source
3x RTX PRO 4000 BlackwellQwen3.5 35B unspecifiedCUDA–3701––655362026-10community–source
3x RTX PRO 4000 BlackwellQwen3.5 35B unspecifiedCUDA–7148––655362026-10community–source

← all models