OpenWeightsTerminal
Pairs

Qwen3.8-Flash-Next

2× AMD MI50 32GB · IQ3_S · Strata

Decode · tok/s
52
Prefill
645
tok/100W
8.7
My Strata result: 2 x AMD MI50 32GB cards, ROCM 10.0.0, Qwen3.8 Flash Next IQ3_S, GPU vision. Prefill: 645 t/s, Decode 52t/s (80t/s with 4 concurrency)

First-person two-card print. Rank is the quoted 52 decode. 80 at 4 concurrency is not a single-stream rank.

Officialclaimed
Open pair