Alibaba's Qwen team posted downloadable Qwen3.8-2.4T-A95B checkpoints on Hugging Face, fulfilling the open-weight promise after the Aug. 3 Qwen3.8-Max API launch. The mixture-of-experts model lists about 2.4 trillion parameters with roughly 95 billion active. Max extras such as vision and default million-token context still sit on the hosted stack, and the license is a custom Qwen grant with commercial triggers. That matters for labs wanting Max-class Chinese weights outside Alibaba Cloud. The caveat is that Qwen3.8-27B was still missing at our check, and serving a 2.4T MoE remains expensive.