Qwen3.6 is the latest addition to the Qwen model family. Building upon the fundamental breakthroughs of Qwen3.5, this release prioritizes stability and real-world utility. The update is aimed at developers seeking a more intuitive, responsive, and productive coding experience, with improvements shaped by direct community feedback. Its main upgrades focus on agentic coding, where the model handles front-end workflows and repository-level reasoning with greater fluency and precision, and thinking preservation, a feature that retains thinking context across conversation history to streamline iterative development and reduce overhead.
Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility. Its enhancements include a unified vision-language foundation trained on trillions of multimodal tokens, an efficient hybrid architecture that combines Gated Delta Networks with sparse Mixture-of-Experts, and scalable reinforcement learning across million-agent environments. Expanded support to 201 languages and dialects enables inclusive, worldwide deployment with nuanced cultural and regional understanding. The training stack also targets near-100% multimodal training efficiency compared to text-only training, with asynchronous reinforcement learning frameworks supporting massive-scale agent scaffolds and environment orchestration.
The release path highlights open availability through Hugging Face Hub and ModelScope. 2026-04-22: Qwen3.6-27B is now available on Hugging Face Hub and ModelScope. 2026-04-16: Qwen3.6-35B-A3B is now available on Hugging Face Hub and ModelScope. 2026-03-02: Qwen3.5-9B, Qwen3.5-4B, Qwen3.5-2B, and Qwen3.5-0.8B are now available on Hugging Face Hub and ModelScope. 2026-02-24: Qwen3.5-122B-A10B, Qwen3.5-35B-A3B, and Qwen3.5-27B are released. Model files can be downloaded automatically through supported frameworks or manually with Hugging Face and ModelScope tooling.
Qwen3.6 can be tried through Qwen Studio, the Qwen API on Alibaba Cloud Model Studio, Qwen Code, and Qwen Agent. Qwen Code is an open-source Artificial Intelligence agent for the terminal, optimized for Qwen models. For local use, Hugging Face Transformers supports inference, training, and serving; llama.cpp supports Qwen3.6 (text & vision); and MLX tools support Qwen3.6 on Apple Silicon. SGLang and vLLM can be used to launch servers with OpenAI-compatible API service, while finetuning is advised through frameworks including UnSloth, Swift, and Llama-Factory. All open-weight models are licensed under Apache 2.0.
