Alibaba has released Qwen3.6-Plus, a new large language model designed to help businesses deploy agentic AI, or systems that can act, decide, and complete tasks with minimal human input.
The model focuses on what Alibaba calls a “capability loop,” meaning it can perceive, reason, and act within a single workflow. This allows it to move beyond simple chatbot responses and handle complex, multi-step tasks such as coding, testing, and refining software.
Qwen3.6-Plus is integrated into Alibaba’s tools, including Wukong, an AI-native enterprise platform, and Qwen App, its flagship AI application. The company said the model is built to support real-world business use, especially as demand grows for AI that can automate entire workflows instead of assisting in isolated steps.
In software development, the model can plan, write, test, and improve code on its own. It supports repository-level engineering, meaning it can handle large, complex codebases. With a context window of up to 1 million tokens, it can process extensive data and instructions in a single session.
Alibaba said the model also improves multimodal reasoning. It can analyze images, documents, and videos, then turn that understanding into actions. For example, it can read user interface screenshots or hand-drawn wireframes and convert them into working frontend code.
“Qwen3.6-Plus is designed to bridge the gap between initial code concepts and deployed products,” the company said.
The model is optimized for accuracy and stability, making it suitable for enterprise use cases such as retail analytics and automated inspections, where consistent results across multiple steps are critical.
Developers can access Qwen3.6-Plus through Model Studio, Alibaba Cloud’s AI development platform, and test it via Qwen Chat. It also works with coding tools like OpenClaw, Claude Code, and Cline to support automated, context-aware development workflows.
Alibaba said it will continue releasing selected Qwen3.6 models to the open-source community in smaller, developer-friendly versions.