Decode · tok/s142Prefill—tok/100W24.7Strataで、Qwen3.8-Flash-Next (IQ3_S) を動かしてみてる。環境はRTX5090と128GB RAM。142 tok/s出てるFirst-person 5090 + 128GB RAM. Separate from the 256K IQ3_S quote at 104. Context not stated.Officialclaimed@arianpgStrataWatchCopy linkOpen pair