On February 16, 2026, the Chinese New Year's Eve, Alibaba released its new generation large model Qwen3.5-Plus, which is comparable to Gemini 3 Pro and has topped the list of the world's strongest open-source large models. This model features a revolutionary underlying architecture, with a total of 397 billion parameters and only 17 billion activated parameters, surpassing the Qwen3-Max with trillions of parameters. The deployment memory usage is reduced by 60%, and the throughput in 256K long context scenarios is increased by 19 times.
It offers excellent cost performance, with API costs at only 0.8 yuan per million tokens, which is 1/18 of that of Gemini 3 Pro. Unlike previous generations that were purely text-based models, it achieves a native multimodal generational leap, pre-trained on mixed visual and text tokens, achieving top performance with less than 40% of the parameters.
It performs exceptionally well in multiple evaluations, scoring 87.8 on MMLU-Pro, surpassing GPT-5.2, achieving 88.4 on GPQA, outperforming Claude 4.5, and breaking records on IFBench. It comprehensively surpasses models such as Gemini 3 Pro and has been synchronized across various Qwen platforms for developers to conveniently access.
