Alibaba Launches Qwen-3.5 in AI Race

Alibaba has unveiled Qwen-3.5, its latest generation AI model family, intensifying the global competition between China and the United States to dominate advanced AI deployment. Released through Alibaba Cloud, the new models mark a strategic push into multimodal, open-weight AI systems designed for large-scale adoption.

The launch, timed ahead of the Lunar New Year, comes amid a wave of flagship releases from Chinese AI developers. Two primary variants—Qwen-3.5-Open-Source and Qwen-3.5-Plus—were made available via Alibaba’s Model Studio platform.

The open-source version features 397 billion parameters and, according to company benchmarks, outperforms the earlier Qwen-3-Max-Thinking model despite the latter exceeding one trillion parameters. Alibaba claims the model performs comparably to leading systems from OpenAI, Anthropic and Google DeepMind, though not necessarily against their newest releases.

A closed-source Qwen-3.5-Plus version offers a 1-million-token context window—among the industry’s largest—allowing it to process vast datasets in a single session.

For the first time, Qwen-3.5 integrates native multimodal capabilities, enabling unified understanding of text, images, audio and video. The models also use Alibaba’s updated architecture to improve computational efficiency and reduce inference costs.

Alibaba released the model weights on developer platforms including Hugging Face and ModelScope, reinforcing China’s growing emphasis on open AI ecosystems. Analysts note that Chinese open models such as Qwen and DeepSeek have recently surpassed US counterparts in download volume, although experts caution that downloads do not equate to real-world deployment.

While Alibaba keeps its largest “Max” models proprietary, it continues bundling open models with cloud services globally—positioning Qwen as both a technological and commercial pillar in the evolving AI race.