Community
Qwen3.8-Flash-Next 125B runs at about 20 tok/s on 16GB VRAM via NVMe offloading
A community project offloads experts from NVMe SSD so the 125B model runs with 8GB system RAM, reaching about 20 tokens per second at 100k to 262k context on an RTX 4090. Benchmarks are self-reported.