ai/qwen3.5 repository overview

⁠Qwen3.5

Qwen3.5 represents a significant advancement in foundation models, delivering exceptional utility and performance through breakthroughs in multimodal learning, architectural efficiency, and global accessibility. The flagship Qwen3.5-397B-A17B model features 397 billion total parameters with 17 billion activated parameters using a sparse Mixture-of-Experts architecture, achieving state-of-the-art results across reasoning, coding, agents, and visual understanding tasks.

This model integrates unified vision-language capabilities through early fusion training on multimodal tokens, achieving cross-generational parity with text-focused Qwen3 models while surpassing previous Qwen3-VL models. The efficient hybrid architecture combines Gated Delta Networks with sparse Mixture-of-Experts to deliver high-throughput inference with minimal latency and cost overhead.

Qwen3.5 provides expanded multilingual support for 201 languages and dialects, enabling inclusive worldwide deployment with nuanced cultural and regional understanding. The model's reinforcement learning approach scales across million-agent environments with progressively complex task distributions for robust real-world adaptability.

⁠Characteristics

⁠Using this model with Docker Model Runner

docker model run qwen3.5

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

⁠Benchmarks

⁠Knowledge

⁠Instruction Following

⁠Long Context

⁠STEM

⁠Reasoning

⁠General Agent

⁠Search Agent

⁠Multilingualism

⁠Links

• Hugging Face Model (GGUF Quantized)⁠
• Qwen Blog - Qwen3.5 Announcement⁠
• Original Model⁠

⁠Considerations

• The model requires substantial computational resources; quantized versions (Q2_K_XL through Q8_0) are available for different hardware configurations
• Native context length is 262K tokens; extended context up to 1M tokens may require additional configuration
• This repository contains GGUF quantized versions optimized by Unsloth for efficient local inference
• The model features a sparse MoE architecture activating only 17B of 397B parameters per token, balancing performance with efficiency
• Multimodal capabilities support both text and image inputs through unified vision-language training
• Vision capabilities require separate mmproj model files (included in repository)

⁠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.