Community
Qwen3.8 Flash Next hits 504 tok/s decode on 2x RTX 5090 with QEngine 1.3.1
A community test reports QEngine 1.3.1 decoding Qwen3.8 Flash Next at 370 to 406 tok/s on the bench and up to 504 tok/s with KV in RAM at long context. Results come from one boot per mode and a repeated-prompt bench, and the author says varied traffic ran about 300 to 400 tok/s.