ai/qwen3.6 repository overview

⁠Qwen3.6

Qwen3.6-35B-A3B is a cutting-edge multimodal AI model from Alibaba Cloud's Qwen team, delivering substantial upgrades in agentic coding, reasoning preservation, and real-world development workflows. Built on direct community feedback following the Qwen3.5 series, this model prioritizes stability and utility, offering developers an intuitive and productive coding experience with enhanced frontend workflows and repository-level reasoning capabilities.

This release features a Mixture of Experts (MoE) architecture with 35 billion total parameters and 3 billion activated parameters, supporting both text and vision inputs. With a native context length of 262,144 tokens (extensible to over 1 million), Qwen3.6 excels at handling complex, multi-turn interactions while maintaining efficiency. The model demonstrates exceptional performance on coding agent benchmarks like SWE-bench Verified, Terminal-Bench, and web development tasks, making it particularly valuable for software engineering, code generation, and multimodal applications.

Key innovations include thinking preservation to retain reasoning context across iterations, improved tool calling with better nested object parsing, and enhanced support for developer roles. Whether you're building coding assistants, multimodal applications, or agentic systems, Qwen3.6 delivers state-of-the-art performance with open-source accessibility.

⁠Characteristics

⁠Using this model with Docker Model Runner

docker model run qwen3.6

For more information, check out the Docker Model Runner docs⁠.

⁠Benchmarks

⁠Coding Agent Benchmarks

⁠General Agent Benchmarks

⁠Knowledge Benchmarks

⁠STEM & Reasoning Benchmarks

⁠Vision Language Benchmarks

⁠Links

• Hugging Face Model (GGUF)⁠
• Qwen Blog: Qwen3.6-35B-A3B⁠
• Official Qwen GitHub⁠
• Qwen Documentation⁠

⁠Considerations

• Hardware Requirements: Due to the MoE architecture with 35B total parameters, running this model efficiently requires adequate GPU memory. GGUF quantized versions are available to reduce memory requirements while maintaining performance.
• Context Window: While the native context length is 262K tokens, optimal performance may vary at extreme lengths. For most applications, the native 262K context provides excellent performance.
• Multimodal Capabilities: The model supports both text and image inputs. Vision capabilities require the appropriate multimodal projection files (mmproj files available in GGUF format).
• Tool Calling: Qwen3.6 includes improved tool calling support with better handling of nested objects. Follow the documented chat template format for optimal tool use.
• Developer Role Support: This model includes enhanced support for developer roles, making it particularly effective for coding assistants and agentic workflows.
• Thinking Preservation: When using the model for iterative development, enable thinking preservation to maintain reasoning context across multiple turns.

⁠Generated by

This model card was automatically generated using cagent-action⁠. Want to learn more about Docker Model Runner? Check out the project repository: https://github.com/docker/model-runner⁠.