- Multimodal model combines reasoning, coding and software execution capabilities
- System can build, test and deploy applications with limited human input
Alibaba Group on June 2 released Qwen3.7-Plus, a new multimodal AI model that can analyze images and video, generate code, use software tools and verify outputs, as the company intensifies competition in the fast-growing AI agent market.
Alibaba said the model ranks among the world’s top five visual AI systems and the highest-ranked Chinese model on major benchmark tests such as Vision Arena.
Qwen3.7-Plus is the latest addition to the Qwen 3.7 family and is designed to combine perception, reasoning and task execution within a single workflow.
App-building capability
The company said the model can process visual inputs, perform multi-step reasoning, write code, invoke external tools and test results before delivering a finished application.
Performance improvements over the previous Qwen3.6-Plus model include gains in coding, agent-related tasks and mathematical reasoning.
Alibaba said the model’s text capabilities approach those of its flagship Qwen3.7-Max model, released two weeks ago.
In one internal test, a system powered by Qwen3.7-Plus spent 11 hours autonomously building an English-learning application.
The process included drafting requirements, writing code, deploying the software, generating test cases, running automated tests and releasing updated versions without human intervention.
The project generated more than 10,000 lines of code and involved thousands of tool calls.
Replicating applications
Alibaba also demonstrated the model’s ability to replicate software applications. In one example, the system analyzed the interface and functionality of Apple’s stock-tracking application, generated its own code, connected to live market data and passed 10 functional verification tests.
The resulting software reproduced features including a dark-mode interface, split-screen layout and real-time market updates.
The model is also designed to bridge software generation and software operation.
Alibaba said users can issue commands such as purchasing a cloud server, after which the system can navigate management consoles, compare configurations, complete purchases and perform operational tasks such as scaling and maintenance.
The launch follows the debut of Qwen3.7-Max in May. While the earlier model focused on advanced reasoning and language capabilities, Qwen3.7-Plus adds visual understanding and task-execution functions that Alibaba sees as critical for next-generation AI agents.
The model is now available through API services on Alibaba Cloud’s Bailian development platform.
