RTX 5090 gpt-oss-120B MXFP4 token-gen @128K
112.0 tok/s TG batch-1 @131K ctx
Fields
id
benchmarks/comm-rtx-5090-gpt-oss-120b-mxfp4-tg-128kMetric
decode_tok_s
Value
112
Unit
tok/s TG batch-1 @131K ctx
Model
gpt-oss-120B
Methodology
llama.cpp, MXFP4 MoE GGUF, entire 120B in 32GB VRAM, batch-1 @131K ctx (Hardware Corner measured table)
Measured by
third_party
Thermal state
unknown
notes
Only consumer card that holds gpt-oss-120B in VRAM. 35B MXFP4 97 t/s @256K.
Notes
Only consumer card that holds gpt-oss-120B in VRAM. 35B MXFP4 97 t/s @256K.