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

User reports running Qwen3.8 Flash-Next through Strata on an RTX 5080 with 64GB system RAM and a Ryzen 7 9800X3D, at roughly 75 t/s with 64K context. Setup uses the Unsloth dynamic GGUF release of the model, with qwen as the agent harness and skills.md files in the repo. The figure is given as approximately 75 t/s, so no exact throughput is recorded. The user says it is slower than Claude but was impressed that the one-shot HTML GTA demo worked without corrections, and notes the model also generated the README and pushed to GitHub.