Alibaba launches Qwen3.7-Plus: a multimodal agent that analyzes video and writes code on its own

Alibaba has released Qwen3.7-Plus, an updated agent model, on its Bailian platform. It understands images and video, can call external tools, and iterates on its own: it writes code, runs it, analyzes errors and rewrites until it reaches the desired result.

What Qwen3.7-Plus is

Qwen3.7-Plus is an intermediate release in the Qwen3 line, tuned for enterprise-grade agent scenarios. In 2026 Alibaba has already released several Qwen3 variants: compact models for local deployment, versions with extended context, and chain-of-thought tooling in Thinking mode. The "Plus" suffix adds multimodality and a full agent stack to the base architecture, while keeping the family's strong position on reasoning and coding benchmarks.

The model is available through Bailian, Alibaba Cloud's AI stack. The platform combines a model repository, RAG tools, an agent-application builder, monitoring and deployment in a single interface. In positioning, Bailian is the closest analog to Amazon Bedrock or Azure AI Studio, aimed at enterprise customers in the Asia-Pacific region.

Four new capabilities

The Qwen team highlights four key updates:

• Vision — the model sees images and video. The agent works with interface screenshots, technical diagrams and video demonstrations without external parsers or separate modal pipelines.
• Deep reasoning — extended step-by-step reasoning for tasks that require chained analysis. The model explains its line of thought, which makes its decisions more transparent and simplifies debugging of agent systems.
• Tool invocation — native calls to external tools, APIs and databases straight from the request context. The model decides for itself when and what to call, without manual orchestration wrappers.
• Autonomous iteration — a self-driven "write → run → analyze → fix" loop. Qwen3.7-Plus does not just generate code: it runs it, reads the error output and rewrites it in cycles until the result is correct.

Autonomous iteration is what sets Qwen3.7-Plus apart from an ordinary chat assistant: the model does not describe actions, it performs them in an environment, reacting to real feedback.

Bailian and the enterprise bet

Launching on Bailian is a deliberate strategic choice. The platform is already integrated with the ERP, CRM and industry databases of Alibaba Cloud's large customers. An agent model with native tool invocation connects to these sources without additional orchestration, lowering the barrier to adoption and shortening the time to first result.

Alibaba positions Bailian as infrastructure for enterprise AI automation. Qwen3.7-Plus becomes the platform's agent core: it sees documents and data, reasons, calls the services it needs and reaches a result through iteration. The bet on its own ecosystem is the key advantage over Western competitors, who find it harder to integrate with the corporate systems of Asian customers because of regulatory barriers and local market specifics.

What this means

The release of Qwen3.7-Plus is part of the broader 2026 trend: AI labs are moving from conversational models to agents capable of autonomously carrying out long tasks in real systems. Alibaba competes with Anthropic, OpenAI and Google by betting not on the model itself but on its integration into enterprise infrastructure. For businesses, this means complex multistep tasks — analyzing documents with images, working with corporate APIs, autonomous coding — become available without hiring a specialized AI team.