The trade dispute between the US and China is having an effect somewhere else. While the US is doing its best to control homegrown technology by imposing tariffs and other measures, tech companies in China are looking to become self-reliant.
Among them is China’s largest retail giant, Alibaba, which already has released its Qwen AI model. From time to time, all these companies upgrade their AI models for better performance and versatility. Well, Alibaba has also done that, as it just released the latest model, Qwen3.5. At a time when agentic AI tools are getting massive attention, the Chinese retail giant is pushing an upgrade in a space that’s already heating up fast.
Qwen3.5 launch brings open-weight and hosted options
Alibaba says that Qwen3.5 comes in two different versions. First up, you have the open-weight edition, which lets you download, run, fine-tune, and deploy the model on your own infrastructure. The hosted version runs directly on Alibaba’s servers, which gives companies a more hands-off option. Both went live on Monday, right before the Chinese New Year. To catch you up, last week, Alibaba also unveiled a new AI model dedicated to robots.
According to the company, Qwen3.5 improves on cost and performance from previous versions. It also comes with “native multimodal capabilities,” which means it can process text, images, and video all in one system. While every major AI model is becoming capable of coding, the latest Qwen model also supports new coding and agentic capabilities and is compatible with open-source agents like OpenClaw. Speaking of OpenClaw, let’s not forget that its founder, Peter Steinberger, has joined OpenAI, with an aim to make personal AI agents accessible to the masses down the line.
Notably, the Qwen3.5 open-weight model comes with 397 billion parameters, slightly fewer than its previous flagship. However, benchmark tests reported by the company hint at solid improvements. It’s worth noting that performance reportedly matches leading models from OpenAI, Anthropic, and Google DeepMind, though these comparisons haven’t been independently verified.
You may also like: Meta’s upcoming ‘Avocado’ AI model could be a part of effort to rebalance its open-source vision
Qwen-3.5-Plus expands capabilities as rivals step up
The hosted variant, Qwen-3.5-Plus, also reportedly delivers competitive results via Alibaba’s Model Studio cloud platform. Another highlight is that the Qwen3.5 AI model now supports 201 languages and dialects, up from 82 in the last generation. Lin Junyang, technical lead of Alibaba Cloud’s Qwen team, said more open-weight models are expected throughout the Chinese New Year.
Meanwhile, Alibaba’s local rivals have also been progressing with their respective AI models. By rivals, here we mean ByteDance and Zhipu AI, which released their upgraded models focusing on agentic capabilities. As the AI race is heating up, Chinese tech giants are doing their every bit to stay in the race and compete aggressively with American counterparts. Not to forget, Alibaba also launched Quark AI Glass series late last year, to compete with Meta & Chinese rivals.
