Qwen3.5 27B
Qwen
Code Multilingual Thinking Tool Calls Vision
Qwen3.5 27B is the largest dense model in Alibaba's Qwen 3.5 medium series, built on the Gated Delta Networks hybrid architecture with 27.78 billion parameters. It scores 86.1 on MMLU-Pro, 85.5 on GPQA Diamond, and 72.4 on SWE-bench Verified, matching GPT-5 mini on coding tasks. 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, extensible to over one million tokens, and covers more than 201 languages. Released under the Apache 2.0 license, it runs in roughly 15 GB of VRAM at Q4, making it an excellent choice for self-hosted deployment on mid-range consumer GPUs.
Hardware Configuration
Optional — for precise deployment recommendations
| Quantization | Quality | Size | Fit |
|---|---|---|---|
| Q8_0 | High | 26.63 GB | — |
| Q8_K_XL | High | 33.09 GB | — |
| Q6_K | High | 20.91 GB | — |
| Q6_K_XL | High | 23.91 GB | — |
| Q5_K_M | Medium | 18.26 GB | — |
| Q5_K_S | Medium | 17.59 GB | — |
| Q5_K_XL | Medium | 18.79 GB | — |
| Q4_K_M | Medium | 15.59 GB | — |
| Q4_K_S | Medium | 14.69 GB | — |
| Q4_K_XL | Medium | 16.41 GB | — |
| Q4_0 | Medium | 14.64 GB | — |
| Q4_1 | Medium | 16 GB | — |
| Q3_K_M | Low | 12.58 GB | — |
| Q3_K_S | Low | 11.45 GB | — |
| Q3_K_XL | Low | 13.45 GB | — |
| Q2_K_XL | Low | 10.44 GB | — |
Last updated: March 13, 2026