Can I run Qwen3.8-Flash-Next on an NVIDIA B300?

Yes, with a GGUF version

The published weights are too big, but the Q8_0 file from unsloth/Qwen3.8-Flash-Next-GGUF (188 GB) fits, with room for ≈ 256K tokens of context.

Full analysis of Qwen3.8-Flash-Next Everything that runs on NVIDIA B300 Get the Q8_0 file on Hugging Face

llama-server -hf unsloth/Qwen3.8-Flash-Next-GGUF:Q8_0

Files and context on NVIDIA B300

Each file of Qwen3.8-Flash-Next (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

File	Size	On NVIDIA B300	Context (16-bit)	Context (8-bit)
As published	360 GB	Doesn't fit	—	—
BF16unsloth/Qwen3.8-Flash-Next-GGUF	354 GB	Doesn't fit	—	—
Q8_0unsloth/Qwen3.8-Flash-Next-GGUF	188 GB	Fits	256K	256K
UD-Q6_K_XLunsloth/Qwen3.8-Flash-Next-GGUF	169 GB	Fits	256K	256K
UD-Q5_K_XLunsloth/Qwen3.8-Flash-Next-GGUF	158 GB	Fits	256K	256K
UD-Q4_K_XLunsloth/Qwen3.8-Flash-Next-GGUF	111 GB	Fits	256K	256K
UD-IQ4_XSunsloth/Qwen3.8-Flash-Next-GGUF	93.7 GB	Fits	256K	256K
UD-Q3_K_XLunsloth/Qwen3.8-Flash-Next-GGUF	89.5 GB	Fits	256K	256K
UD-IQ3_XXSunsloth/Qwen3.8-Flash-Next-GGUF	83.6 GB	Fits	256K	256K
UD-Q2_K_XLunsloth/Qwen3.8-Flash-Next-GGUF	79.6 GB	Fits	256K	256K
UD-IQ2_XSunsloth/Qwen3.8-Flash-Next-GGUF	75.8 GB	Fits	256K	256K
Qwen3.8-Flash-Next-UD-Q2_0unsloth/Qwen3.8-Flash-Next-GGUF	70.5 GB	Fits	256K	256K
UD-IQ1_Munsloth/Qwen3.8-Flash-Next-GGUF	68.0 GB	Fits	256K	256K
UD-IQ1_Sunsloth/Qwen3.8-Flash-Next-GGUF	66.4 GB	Fits	256K	256K
Q5_K_MAtomicChat/Qwen3.8-Flash-Next-GGUF	111 GB	Fits	256K	256K
Q4_K_MAtomicChat/Qwen3.8-Flash-Next-GGUF	94.5 GB	Fits	256K	256K
IQ4_XSAtomicChat/Qwen3.8-Flash-Next-GGUF	84.9 GB	Fits	256K	256K
Q8_0ggml-org/Qwen3.8-Flash-Next-GGUF	163 GB	Fits	256K	256K
IQ4_NLggml-org/Qwen3.8-Flash-Next-GGUF	102 GB	Fits	256K	256K

Qwen3.8-Flash-Next on other hardware

Popular models that run on NVIDIA B300

Assumes 90% of the card's memory is usable and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.