Qwen3.6-27B-GGUF

Qwen3.6-27B from Alibaba's Qwen team is the latest dense multimodal language model in the Qwen3.6 series, featuring 27 billion parameters with a hybrid Gated DeltaNet architecture (64 layers, hidden dim 5120, 248K vocabulary across 201+ languages) and an integrated vision encoder. Specifically optimized for agentic coding, repository-level reasoning, and stable real-world utility, it supports multi-token prediction and a native context length of 262,144 tokens—extensible up to 1M+ tokens via YaRN. The model introduces "Thinking Preservation" to maintain reasoning context across iterative developments and delivers flagship-level performance across various benchmarks. It notably achieves 77.2 on SWE-bench Verified, 83.9 on LiveCodeBench v6, and 94.1 on AIME26 for language and coding tasks, while its multimodal strengths are evidenced by high scores in vision-language evaluations, including 82.9 on MMMU, 87.4 on MathVista mini, and 87.7 on VideoMME, making it a highly capable tool for both complex programming and sophisticated visual understanding.

Model Files

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):

👁 image.png

Base model