Alibaba Unveils Latest AI Model Qwen 2.5-Max, Challenges DeepSeek Rising Influence

The Chinese giant of Alibaba technology has launched its latest IA model, Qwen 2.5-Max, in a challenge to the growing influence of Deepseek.

The deployment that took place on the first day of the lunar New Year, highlights Alibaba’s desire to remain competitive in the middle of the rapid Deepseek ascent.

In an effort to reaffirm his domination, Alibaba shared reference results suggesting that Queen 2.5-Max surpasses the models of Openai and Anthropic in certain tasks. The company said that its IA model achieves competitive performance with high-level models, referring to GPT-4O, Deepseek-V3 of Openai, and Llama-3.1 of Meta, on the basis of its compilted comparison the help of various reference tests.

Alibaba also announced that Qwen2.5-Max is now available in Qwen Chat, her chatbot Web interface where users can interact directly with the model. According to Alibaba Cloud, Qwen 2.5-Max, according to Alibaba Cloud, shows that the expansion of the data scale and the model parameters can effectively improve the intelligence of an AI model.

Qwen 2.5 consists of dense language models and decoder ranging from 0.5 billion to 72 billion parameters, available in basic and educated variants. The model is pre-formed on a set of 18 billion data, supporting multilingual capacities in 29 languages, including Chinese, English, Spanish and Arabic. It can treat up to 128,000 tokens in context and generate outings up to 8,000 tokens.

In addition to linguistic capacities, Qwen 2.5 improves documents analysis, object detection and understanding video. It improves precision in the identification of objects through several formats. In addition, Qwen 2.5 is delivered with a better video understanding, taking a step on understanding the image of competition thanks to an ultra-long video processing and a fine-grained video setting.

Here are several other features that bring out Qwen:

Omni-modal mastery

Can analyze documents (tables, graphics, writing), analyze long videos and generate structured outputs (invoices, forms, etc.).

Training efficiency

PRETRABLE ON 18 to 20 Billions of tokens with multimodal learning and fine adjustment for advanced performance.

Reference domination

GPPT -4 and Deepseek surpasse – saying high -level results with 40 to 60% less resources.

In particular, this model applies a dynamic resolution and training at the frequency of images for a better temporal understanding, helping it to extract event segments effectively. It also incorporates a rationalized vision encoder which improves both Training and inference speeds.

The deployment of Qwen2.5 by Alibaba follows a market frenzy triggered by the Chinese startup in depth, in the midst of concerns concerning demand for the most advanced chips and data centers. Industry observers believe that Alibaba’s announcement reflects the pressure that the Deepseek IA model has not only on world competitors but on internal competition.

The release of the AI ​​assistant from Deepseek, fueled by the Deepseek-V3 model, also shocked Silicon Valley and plunged technological shares, with Chinese startups with Chinese startups Allegedly low development and use costs Undressing investors to question huge spending plans by directing AI companies in the United States.

In particular, the rise of Deepseek-V3 influenced the trends in the AL market, leading to upgrades of Alibaba, Bytedance and Tencent. The Open-Source startup model would have launched an AL price war in China, which caused costs Reductions between competitors. The latest Alibaba model aims to provide an alternative with advanced multimodal functions and optimized inference speeds.

With the Chinese technology giants that embark on AI, the rapid development of AL models in the country, means the growing importance of technology. While the competition warms the world technological ecosystem, we can expect new innovations and pierced in the field of AL.