Alibaba Launches Qwen3.5 to Power Smarter AI Agents

Alibaba has unveiled Qwen3.5, a new family of AI models designed to move beyond basic chatbots and into autonomous task execution. The company is positioning the system as a major step toward AI that can plan, reason, and complete real-world workflows with minimal human input.

The release includes both open-weight models for developers and a hosted cloud version, signaling Alibaba’s push to compete with leading global AI platforms.

What Qwen3.5 Actually Is

Qwen3.5 is Alibaba’s latest large AI model series built specifically for agent-style tasks. Instead of just answering questions, the system is designed to help software agents carry out complex, multi-step actions.

Alibaba is offering the model in two main ways:

• Open-weight version: Developers can download and self-host
• Qwen3.5-Plus: A managed version available via Alibaba Cloud Model Studio

This gives companies flexibility depending on whether they want full control or a ready-to-use cloud service.

Built as a Native Multimodal Model

Alibaba says Qwen3.5 is natively multimodal, meaning it can handle text, images, and video within one unified system.

This is important because modern AI agents often need to:

• Read screenshots
• Understand visual interfaces
• Process documents
• Analyze video content

By combining these capabilities, Alibaba is aiming to make Qwen3.5 more useful for real automation tasks.

Key Technical Details

The flagship Mixture-of-Experts version, often called Qwen3.5-397B, includes:

• 397 billion total parameters
• About 17 billion active per token
• Support for very long context windows
• Up to ~1 million tokens in the Plus version

The MoE design helps the model stay powerful while keeping inference costs lower than dense models of similar size.

Performance and Cost Claims

Alibaba is making strong efficiency claims with this release.

According to the company:

• Qwen3.5 is around 60 percent cheaper to run than the previous version
• It delivers about 8× higher throughput on large workloads
• Internal tests show performance on par with top Western models

However, these benchmark results are self-reported and have not yet been independently verified.

Strong Focus on AI Agents

One of the biggest upgrades is support for agent workflows.

Qwen3.5 is optimized for:

• Function calling
• Multi-step planning
• External tool connections (via MCP and similar protocols)
• Long-context reasoning

Alibaba is also highlighting visual agent capabilities, such as:

• Reading UI screenshots
• Generating matching HTML and CSS
• Controlling desktop or mobile apps

In benchmarks like IFBench, the model reportedly scores in the mid-70s, suggesting competitive performance for complex agent tasks.

How It Fits Into Alibaba’s Ecosystem

Qwen3.5 powers the upgraded Qwen App, which Alibaba is turning into a full-service AI assistant.

The assistant can help users:

• Order food
• Book travel
• Shop on Taobao
• Pay through Alipay
• Navigate using Amap
• Coordinate services via Fliggy

Alibaba describes this shift as moving from “AI that responds” to “AI that acts.”

The new Task Assistant feature is designed to handle multi-step workflows like searching, comparing, booking, and paying in one conversation.

Growing Competition in China’s AI Race

The launch comes during an intense agentic AI race in China.

Key competitors include:

• ByteDance’s Doubao 2.0
• Zhipu AI’s latest models
• DeepSeek’s rapidly rising global profile

Doubao reportedly has close to 200 million users, while DeepSeek has gained significant international attention.

Qwen3.5 is clearly part of Alibaba’s effort to stay competitive in this fast-moving market.

Why the Open-Weight Strategy Matters

By releasing open weights alongside a cloud version, Alibaba is trying to attract a broad developer base.

Benefits for developers include:

• Greater deployment control
• Better cost management
• Improved data privacy options
• More flexibility for custom AI agents

This hybrid approach could help Alibaba expand beyond its domestic ecosystem.

What to Watch Next

Qwen3.5 looks ambitious, but several questions remain.

Industry observers will be watching for:

• Independent benchmark validation
• Real-world agent reliability
• Enterprise adoption outside China
• Developer traction globally

These factors will determine how competitive the model becomes internationally.

Bottom Line

Alibaba’s Qwen3.5 marks a clear shift toward AI systems that can take action, not just generate text. With multimodal support, strong agent features, and an open-weight strategy, the company is positioning itself for the next phase of the AI platform race.

The real test will come as developers and enterprises begin using the model in production environments.