Qwen

Qwen3.8 Max (2.4T)

Qwen's 2.4T-parameter (95B active) MoE flagship; native context 262,144 tokens, extensible to about 1.01M.text generationother26.3K downloadsHugging Face

No server in stock right now has the 5871 GB this model needs for inference.

License restrictions

Model-as-a-Service or AI Work Assistant operators with over USD 50M revenue in 12 months need a separate license; products above 100M MAU or USD 20M monthly revenue must display the model name.

GPU memory it needs

From model cards and vendor documentation. Lower precision needs less memory, with slightly lower quality. Estimates are marked.

What you want to do	Precision	Minimum	Recommended	Setup	Basis
Run the model vLLM recipe minimum for this repo's 4.45 TiB BF16 checkpoint.	BF16, full	5871 GB	6432 GB	24 GPUs: 3x 8-GPU B300 nodes or 6 GB300 NVL4 trays	Published
Run the model vLLM recipe minimum for separate repo Qwen/Qwen3.8-2.4T-A95B-FP8 (2.27 TiB).	FP8, 8-bit	2996 GB	4288 GB	16 GPUs: 2x 8-GPU B300 nodes or 4 GB300 NVL4 trays	Published
Run the model vLLM recipe minimum for separate repo Inferact/Qwen3.8-2.4T-A95B-NVFP4 (1.32 TiB), Blackwell only.	4-bit	1737 GB	2144 GB	8 GPUs: 1x 8-GPU B300 node or 2x GB300 NVL4 trays	Published