2 min readNew DelhiMar 4, 2026 01:43 PM IST AI companies around the world are constantly working on newer, larger and more powerful models, but Chinese tech giant Alibaba seems to be going the opposite way. Recently, the company unveiled four new small models under its Qwen 3.5 series.
These open-source large language models, including Qwen 3.5-0.8B, 2B, 4B, and 9B, are aimed at developers seeking multimodal capabilities alongside strong reasoning performance.
Alibaba says all four models in the Qwen 3.5 series are built on the same architecture and can easily handle both text and images. For every model in the series, the company has released a ‘base’ version that developers can fine-tune from scratch, along with an ‘instruct’ version that can be deployed immediately.
However, the largest and most powerful in the series – Qwen 3.5-9B is getting the most attention. In terms of benchmarks, Alibaba says the large language model performs similarly to the GPT-oSS-120B despite the size difference between the two. The Chinese tech giant also claims that the model’s performance is comparable to AI chatbots like ChatGPT and Gemini in several reasoning and knowledge-based tests. In tasks that involve logical reasoning, math problem-solving, and document analysis, the Qwen 3.5-9B performs similarly to these much larger AI chatbots.
On the other side of the spectrum, the Qwen 3.5-0.8B and 2B, two lightweight models that are designed to run on portable devices like laptops and smartphones, where computing power is limited at best. And while they are not as strong as other Qwen models in reasoning, Alibaba says they can easily process both text and images.
The new Qwen 3.5 series models are available on Hugging Face as well as ModelScope with open weights, and can be downloaded and run locally using popular AI frameworks. Responding to a tweet, xAI head Elon Musk has said that the new Qwen 3.5 series models have “impressive intelligence density.”
Expand © IE Online Media Services Pvt Ltd
