Skip to content

Qwen3.5 397B A17B

Qwen
Code Multilingual Thinking Tool Calls Vision

Qwen3.5 397B A17B is the largest Mixture-of-Experts model from Alibaba's Qwen team with 403 billion total parameters and only 17 billion active per token, routed across 512 experts for exceptional efficiency at scale. It is natively multimodal, processing text, images, and video, with built-in thinking capabilities for chain-of-thought reasoning. The model supports a 262K context window and covers over 200 languages. Released under the Apache 2.0 license, it delivers frontier-level performance on math, code, and vision benchmarks, making it a compelling self-hosted alternative for demanding workloads.

Hardware Configuration

Optional — for precise deployment recommendations
Quantization Quality Size Fit
BF16 Full precision 738.5 GB
MXFP4_MOE Very high 221.04 GB
Q8_0 High 392.55 GB
Q8_K_XL High 398.37 GB
Q6_K High 304.17 GB
Q6_K_XL High 337.44 GB
Q5_K_M Medium 273.49 GB
Q5_K_S Medium 257.57 GB
Q5_K_XL Medium 274.61 GB
Q4_K_M Medium 227.33 GB
Q4_K_S Medium 212.33 GB
Q4_K_XL Medium 228.44 GB
IQ4_NL Medium 180.45 GB
IQ4_XS Medium 176.7 GB
Q3_K_M Low 165.22 GB
Q3_K_S Low 153.03 GB
Q3_K_XL Low 166.46 GB
IQ3_S Low 136.32 GB
IQ3_XXS Low 130.69 GB
IQ2_M Low 114.61 GB
IQ2_XXS Low 106.99 GB
IQ1_M Low 99.49 GB
Last updated: March 20, 2026