Alibaba Cloud officially launched Qwen 3.5, a next-generation AI model family designed specifically for the “agentic AI” era. The release marks a major escalation in the global AI race, with Alibaba claiming its new flagship outperforms Western rivals like GPT-5.2 and Claude Opus 4.5 across several key benchmarks.
The centerpiece of the launch is the Qwen3.5-397B-A17B, a massive open-weight model that uses a sparse Mixture-of-Experts (MoE) architecture to deliver high-tier intelligence with extreme efficiency.
Key Capabilities and Innovations
Qwen 3.5 is engineered to act as an autonomous agent rather than a simple chatbot, featuring what Alibaba calls “visual agentic capabilities.”
- Autonomous Execution: The model can independently perform tasks across mobile and desktop applications, such as navigating software interfaces, managing schedules, or executing complex multi-step workflows.
- Efficiency Breakthrough: Despite its 397-billion-parameter size, it activates only 17 billion parameters per forward pass. This makes it 60% cheaper to operate and 8 times more efficient than its predecessor, Qwen 2.5.
- Native Multimodality: It is a native vision-language model, capable of understanding and analyzing up to 2 hours of video and processing complex documents with high layout fidelity.
- Global Reach: Language support has been expanded from 119 to 201 languages and dialects, specifically optimized for Southeast Asian and regional Indian scripts.
Benchmark Performance
In results published alongside the launch, Qwen 3.5 demonstrates striking competitive strength against the latest global frontier models:
| Benchmark | Category | Qwen 3.5 | GPT-5.2 | Claude 4.5 Opus |
| IFBench | Instruction Following | 76.5 | 75.4 | 58.0 |
| GPQA Diamond | Graduate Reasoning | 88.4 | 92.4 | 87.0 |
| BFCL-V4 | Agentic Tool Use | 72.9 | 63.1 | 77.5 |
| Video-MME | Video Reasoning | 87.5 | — | — |
The Qwen 3.5 Family
Alibaba is offering two primary ways to access the new intelligence:
- Qwen3.5-397B-A17B (Open Weight): Available on Hugging Face and ModelScope for developers to download and deploy. It features a 256k context window.
- Qwen3.5-Plus (Hosted): Available via Alibaba Cloud Model Studio. It comes with a massive 1 million-token context window by default and supports “adaptive tool use” for complex enterprise automation.
Strategic Market Context
The launch arrived just one day after ByteDance released Doubao 2.0, signaling a fierce “agentic” rivalry in China. Alibaba is leveraging this launch to convert its recent massive user surge—which saw its app activity jump seven-fold following a coupon campaign—into long-term enterprise adoption.
“Built for the agentic AI era, Qwen 3.5 is designed to help developers and enterprises move faster and do more with the same compute, setting a new benchmark for capability per unit of inference cost.” — Alibaba Cloud Statement.
