List your company
ResearchFormerly WizWand

Prefill Throughput on Synthetic

20,864.85Prefill Throughput (tokens/s)YOCO (Dense)
225.22845,583.591710,941.95516,300.3183Jun 4, 2026
Updated 4mo ago

Evaluation results

MethodLinks
YOCO (Dense)Context Length=128K, Hardware=NVIDIA B200 GPUs2026.0620,864.85
YOCO (CLSA)Context Length=128K, Hardware=NVIDIA B200 GPUs2026.0620,741.51
YOCO (CLSA)Context Length=64K, Hardware=NVIDIA B200 GPUs2026.0620,349.11
YOCO (Dense)Context Length=64K, Hardware=NVIDIA B200 GPUs2026.0620,343.94
YOCO (Dense)Context Length=32K, Hardware=NVIDIA B200 GPUs2026.0618,450.18
YOCO (CLSA)Context Length=32K, Hardware=NVIDIA B200 GPUs2026.0618,163.24
YOCO (Dense)Context Length=16K, Hardware=NVIDIA B200 GPUs2026.0613,033.56
YOCO (CLSA)Context Length=16K, Hardware=NVIDIA B200 GPUs2026.0611,572.4
YOCO (Dense)Context Length=8K, Hardware=NVIDIA B200 GPUs2026.0610,884.35
YOCO (CLSA)Context Length=8K, Hardware=NVIDIA B200 GPUs2026.069,623.48
TransformerContext Length=8K, Hardware=NVIDIA B200 GPUs2026.064,721.71
TransformerContext Length=16K, Hardware=NVIDIA B200 GPUs2026.064,160.92
TransformerContext Length=32K, Hardware=NVIDIA B200 GPUs2026.062,889.39
TransformerContext Length=64K, Hardware=NVIDIA B200 GPUs2026.061,741.56
TransformerContext Length=128K, Hardware=NVIDIA B200 GPUs2026.061,019.06