Qwen3.7-Plus: Multimodal Agentic Intelligence

Qwen3.7-Plus is a multimodal agentic foundation model that unifies vision and language into a single model. Building on the strong text performance of the existing Qwen3.7, it significantly strengthens vision-language capabilities while maintaining powerful agentic capabilities across coding, tool use, and productivity workflows.

The model can perceive real-world scenes, read screens, and operate GUIs. It can write code based on visual references or navigate mobile apps end-to-end, and it combines web knowledge to answer visual questions. In particular, its core strength lies in seamlessly combining GUI and CLI interactions within a single agent loop.

Key capabilities include:

• Multimodal interactive hybrid agent: unified GUI/CLI operation spanning both visual and text tasks
• General-purpose coding agent: supports everything from frontend prototyping to complex software engineering
• Visual agent: performs perception, reasoning, grounding, and retrieval-augmented QA
• Framework versatility: delivers consistent performance across various agent scaffolds such as Claude Code, OpenClaw, and Qwen Code

It is currently available via API through Alibaba Cloud Model Studio. Benchmark results demonstrate outstanding performance surpassing competing models across various agentic and reasoning metrics, including Terminal Bench 2.0-Terminus (70.3) and QwenWorldBench (62.1).