Pair

Qwen3.8-Flash-Next

Name: OpenWeights Terminal measured pairs
Creator: Dvidia
License: https://owterminal.com/cite

One quoted capture. Rank only lives inside a filter.

2× RTX 3060 12GB · IQ3_S · Strata

Decode · tok/s

Prefill

—

tok/100W

11.8

短コンテキストだと 40 tok/s、50k 超の長コンテキストで 10% 程度減速するけどほぼ誤差の範囲。モデルは公式版の IQ3_S

First-person 2× 3060 12GB. Rank is the quoted 40 short-context decode. Prefill was not in this post.

Officialclaimed