On June 2, 2026, Alibaba's Tongyi Qianwen team officially released the Qwen3.7-Plus multimodal large model. The model is a multimodal upgrade of the Qwen3.7 series, positioned as a unified vision-and-language agent foundation, realizing an end-to-end closed loop of “see, think, write, act, verify”. Qwen3.7-Plus is now offered via the Alibaba Cloud Bailian platform with API access and a limited-time 20% discount on postpaid inference.

Building on Qwen3.7's strong text, code and tool-usage capabilities, Qwen3.7-Plus significantly enhances visual understanding and multimodal interaction. It can process images and video inputs, and additionally supports screenshots, GUI understanding and real-world operations, becoming the first domestically available open-model-level system to truly offer multimodal hybrid agent capabilities.

Key Capability Breakthroughs

According to Alibaba's technical report and third-party evaluations, Qwen3.7-Plus has achieved significant improvements on multimodal tasks:

Visual understanding and interface operation: The model can accurately interpret mobile/desktop screenshots, design mockups or real-scene photos, and autonomously generate operation commands, code or execution steps. For example, given an app interface screenshot, the model can produce corresponding front-end code or automated operation scripts.
Multimodal agent closed loop: It supports the full workflow of “view→reason→write code→invoke tools→verify results,” making it well suited for complex GUI automation, desktop software replication and multi-step agent tasks.
Benchmark performance: It ranks among the top five on the Vision Arena global visual large-model leaderboard and first in China; it also shows leading performance on multimodal benchmarks such as ScreenSpot Pro and OmniDocBench.

Compared with the previous Qwen3.7 generation, Qwen3.7-Plus realizes a qualitative leap in vision–language fusion and long-context multimodal reasoning while maintaining relatively low inference costs, making it suitable for enterprise-scale deployment.

Technical Architecture Highlights

Qwen3.7-Plus adopts a native multimodal hybrid architecture rather than a post-hoc concatenated visual module. This enables more natural and efficient cross-modal processing. Key features include:

A unified vision–language encoder supporting high-resolution images and long videos;
An enhanced agentic reasoning chain that supports long-duration, multi-tool collaborative tasks;
An optimized context management mechanism with default support for million-token-scale long inputs, suitable for complex document analysis and extended agent workflows.

The model is already available on the Alibaba Cloud Bailian platform; users can invoke it with a single line of code, submit text, image or video inputs, and integrate it with mainstream agent frameworks.

Industry Significance and Impact

The release of Qwen3.7-Plus further strengthens China's competitiveness in multimodal and agent-capable large models. It provides developers with a high-performance, cost-effective multimodal toolset and serves as an important foundation for deploying AI agents in production environments.

Accelerating multimodal agent adoption: Breakthroughs in GUI understanding and operation are expected to drive fast development of applications such as automated testing, desktop assistants and intelligent office tools.
Clear cost-performance advantage: Compared with top-tier closed-source international models, Qwen3.7-Plus delivers high performance with notable cost advantages, encouraging broader adoption by SMEs and developers.
Boosting the open-source ecosystem: Alibaba Cloud has also open-sourced