Qwen

Qwen3.8 Flash Next

Qwen4-architecture preview MoE (about 176B total, 6B active parameters) with image input and native 262,144-token context, extensible to 1M.image text to textother3.5M downloadsHugging Face

No server in stock right now has the 423 GB this model needs for inference.

License restrictions

Model-as-a-Service or AI Work Assistant operators need a separate license for commercial use (internal use is exempt); 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 BF16 checkpoint; recipe default command uses the NVFP4 variant instead.	BF16, full	423 GB	—	—	Published
Run the model vLLM recipe minimum for separate repo Qwen/Qwen3.8-Flash-Next-FP8.	FP8, 8-bit	250 GB	—	—	Published
Run the model vLLM recipe minimum for separate repo Inferact/Qwen3.8-Flash-Next-NVFP4, FP8 KV cache.	4-bit	130 GB	1440 GB	8x B200 (vLLM recipe default for the NVFP4 variant)	Published