- Alibaba has launched next-generation large model, vying for global open-source leadership while joining the battle for holiday users.
- The latest iteration of its Qwen model, Qwen3.5, is designed to support AI agent tasks and can understand text, image, and video inputs.
Alibaba unveiled a major upgrade to its flagship AI model on Monday, accelerating its race against numerous startups and industry leaders — all vying for an edge ahead of DeepSeek's next major platform launch in China's AI sector.
In a statement, Alibaba said the latest iteration of its Qwen model, Qwen3.5, is designed to support AI agent tasks and can process text, image, and video inputs.
The company added that the model can analyze video content up to two hours in length.
Alibaba quietly launched two new models — Qwen3.5-Plus and Qwen3.5-397B-A17B — on its chat.qwen.ai page Monday afternoon.
Qwen3.5-Plus is positioned as the latest large language model in the Qwen 3.5 series, while Qwen3.5-397B-A17B serves as the flagship open-source large language model in the Qwen3.5 series.
Alibaba said that Qwen3.5 achieves a comprehensive overhaul of its underlying model architecture. The Qwen3.5-Plus version boasts a total of 397 billion parameters, yet requires only 17 billion parameters to activate at any given time, outperforming the Qwen3-Max model with over a trillion parameters.
The innovation reduces deployed memory usage by 60%, significantly enhances inference efficiency, and increases maximum inference throughput by up to 19 times.
The model employs an innovative hybrid architecture combining linear attention (Gated Delta Networks) with sparse mixture-of-experts (MoE).
At 32k/256k context lengths, Qwen3.5-397B-A17B achieves decoding throughput 8.6 times and 19.0 times faster than Qwen3-Max, respectively.
The Qwen3.5 release is part of a wave of upgrades ahead of the Lunar New Year. From ByteDance to Zhipu AI and Minimax, companies have intensively launched new versions in recent weeks, aiming to gain an edge before DeepSeek's highly anticipated product release.
DeepSeek is expected to unveil major updates around the one-year anniversary of its disruptive R1 model, which shook the global AI industry in 2025.
These companies are also targeting the week-long Spring Festival holiday — a period historically proven critical for the adoption of Chinese consumer internet applications.
Since DeepSeek reactivated the domestic tech sector, Alibaba has emerged as one of the most aggressive investors and advocates in AI. CEO Eddie Wu has pledged over $53 billion for infrastructure and AI development — a commitment he indicated could grow further over time.
Alibaba last upgraded its flagship model to the Qwen3 series in April of last year. Last month, the company launched its latest inference model, Qwen3-Max-Thinking, designed to enhance performance in areas like complex reasoning and autonomous agent capabilities.
