Qwen Announces Launch of Latest Open Source Dense Model Qwen3.6-27B, a 27 Billion Parameter Multimodal Model

Qwen announced the launch of its latest open source dense model Qwen3.6-27B, which is a 27 billion parameter multimodal model. It outperforms the previous 397 billion parameter MoE model Qwen3.5-397B-A17B across major programming benchmarks such as SWE-bench, Terminal-Bench, and SkillsBench, and is officially positioned as an open source solution with "flagship programming capabilities." According to the official introduction with several English tech media, this model is also highly competitive in text and multimodal reasoning tasks, achieving results close to top proprietary models in challenging reasoning tests like GPQA.

Qwen3.6-27B adopts a dense architecture, supporting both thinking and non-thinking modes, with visible chain reasoning content enabled by default, and offers a "retain reasoning" option for continuous reuse of intermediate thoughts in multi-turn development scenarios. Users can also configure API parameters to disable thinking output, retaining only direct answers. The model is fully open source under the Apache 2.0 license, with weights available for download and inference services provided on Hugging Face and Qwen Studio, along with multiple GGUF quantization versions (3–6 bits), allowing it to run on local hardware with approximately 15–30 GB, lowering the deployment threshold for self-built environments and privacy-sensitive code assistants.

Source: Public Information