Qwen3.8-Flash-Next

One quoted capture. Rank only lives inside a filter.

RTX 5090 32GB · IQ3_S · Strata

Decode

104

Prefill

1657

tok/100W

18.1

Confirmed on RTX 5090 (IQ3_S, 256K ctx): prefill 1221→1657 tok/s (+36%), decode 63→104 tok/s (+65%). Full expert profile filled VRAM 22.5→31GB.

Reply to Niko Veit Strata kernel bump. Rank uses the after numbers (104 decode / 1657 prefill), not the before arrow.

Officialclaimed