Qwen3.8 125B (6B active) Flash-Next

User reports Qwen3.8 Flash-Next at about 50 t/s on an RTX 5080 with 64GB system RAM, down from part 1 because the context window was raised from 64K to 128K. Setup is Strata with an Unsloth UD-IQ4_XS GGUF at 128K context, driven by the Qwen agent harness. A second model, Qwen3.6-35B-A3B UD-Q4_K_S, runs on an M1 Max MacBook with 64GB at about 550 t/s for fast tasks, two parallel requests. The game build took 27 hours start to finish.