Skip to content

Qwen3 Coder Next

Qwen
Code Multilingual Tool Calls

Qwen3 Coder Next is a code-specialized Mixture-of-Experts model from Alibaba's Qwen team with 79.67 billion total parameters, purpose-built for agentic coding and tool-use workflows. Only around 3 billion parameters activate per token, activating 10 of 512 experts, achieving code performance comparable to far larger models at a fraction of the compute. It supports tool calling and 13 languages including English and Chinese. With a 262K context window and flash attention, it handles large codebases natively and quantizes well to GGUF for self-hosted development environments.

Hardware Configuration

Optional — for precise deployment recommendations
Quantization Quality Size Fit
MXFP4_MOE Very high 44.73 GB
Q8_0 High 78.99 GB
Q8_K_XL High 79.84 GB
Q6_K High 61.04 GB
Q6_K_XL High 61.28 GB
Q5_K_M Medium 52.91 GB
Q5_K_S Medium 51.25 GB
Q5_K_XL Medium 52.97 GB
Q4_K_M Medium 45.18 GB
Q4_K_S Medium 42.38 GB
Q4_K_XL Medium 45.18 GB
Q4_0 Medium 42.01 GB
Q4_1 Medium 46.62 GB
Q3_K_M Low 35.66 GB
Q3_K_S Low 32.21 GB
Q3_K_XL Low 35.78 GB
Q2_K Low 27.15 GB
Q2_K_L Low 27.22 GB
Q2_K_XL Low 27.49 GB
Last updated: March 12, 2026