Aliyun Releases Tongyi Qianwen 2.5 Model: Enhanced Capabilities Outshine GPT-4

During yesterday's Aliyun AI Leaders Summit in Beijing, Aliyun officially launched the Tongyi Qianwen 2.5 model, claiming that it surpasses GPT-4 in several capabilities, particularly in understanding and generating content in Chinese.

According to Aliyun, since the introduction of the Tongyi model, it has served over 90,000 enterprises, with open-source model downloads exceeding 7 million. The application of this model is rapidly expanding across various sectors, including PC, mobile, and automotive.

As the latest addition to the Tongyi model series, Tongyi Qianwen 2.5 shows significant advancements in text comprehension, generation, question answering, lifestyle recommendations, casual conversation, and risk management. Aliyun reported improvements of 9% in understanding capabilities, 16% in logical reasoning, 19% in instruction compliance, and 10% in coding capabilities.

The model is open-source with 110 billion parameters and has demonstrated outstanding performance in multiple benchmark tests, outperforming Meta's Llama-3-70B, making it one of the most powerful models in the open-source domain.

At the summit, Xiaomi's AI assistant, Xiao Ai, announced a partnership with Aliyun's Tongyi model, paving the way for deep integration and upgrade of multimodal capabilities in operating systems. Additionally, companies like Weibo and Perfect World Games have plans to adopt the Tongyi model to drive innovation and development in AI technology.

The launch of Tongyi Qianwen 2.5 showcases Aliyun's continued advancements in the AI field and underscores its commitment to innovating and applying AI technologies, offering more efficient and intelligent solutions across various industries.