Alibaba Cloud Launches Qwen2 Series LLM

Alibaba Cloud Launches Qwen2 Series LLM

Kai Chen
2 min read

Alibaba Cloud Launches Qwen2 Series Model with up to 720 Billion Parameters

Alibaba Cloud has unveiled the Qwen2 series model on June 7th, featuring five versions with parameter counts ranging from 500 million to an impressive 720 billion. Among them, the Qwen2-72B model has surpassed Meta's Llama3-70B in various evaluations, signifying a significant advancement in large model technology. The Qwen2-57B, as Alibaba Cloud's second hybrid expert model (MoE), offers enhanced performance within the same resource constraints, portraying it as a new trend in big model technology. Over the past year, Alibaba Cloud has been actively promoting the development of open-source models in China, with the release of the Qwen2 series further solidifying its leading position in the open-source domain. Moreover, the Qwen2-72B has demonstrated comprehensive superiority in evaluations of common sense, logical reasoning, and mathematical capabilities, showcasing its exceptional performance across multiple crucial domains.

Key Takeaways

  • Alibaba Cloud introduces the Qwen2 series model, encompassing five versions with parameter counts ranging from 500 million to 720 billion.
  • The Qwen2-72B model outperforms Meta's Llama3-70B in multiple evaluations.
  • Qwen2-57B marks Alibaba Cloud's second hybrid expert model (MoE) with enhanced performance.
  • Alibaba Cloud has been actively promoting the development of open-source models within China over the past year.
  • Open-source model technology is deemed a key driver in the development of large models.


The introduction of Alibaba Cloud's Qwen2 series models, particularly the Qwen2-72B and Qwen2-57B, marks a significant leap forward in large model technology. The superior performance of Qwen2-72B compared to Meta's Llama3-70B not only enhances Alibaba Cloud's competitive edge in the open-source domain but also has the potential to impact the global market landscape of AI technology. As an MoE model, the high efficiency of Qwen2-57B foreshadows a new direction in AI model design. In the short term, this technological breakthrough may attract more enterprises and research institutions to adopt Alibaba Cloud's services, while in the long term, it could drive technological innovation and efficiency improvements across the entire AI industry. Furthermore, Alibaba Cloud's open-source strategy facilitates the creation of a broader ecological system, bolstering its influence in the global market.

Did You Know?

  • Mixture of Experts (MoE): MoE is a neural network architecture that uses multiple "experts" within the network to handle different types of data or tasks. Each expert specializes in specific tasks or data types within the network, allowing MoE to provide stronger performance and flexibility while maintaining computational efficiency. This architecture is particularly suitable for handling large-scale and complex models, as it enables the allocation of tasks among different experts to optimize resource utilization and enhance performance.
  • Open-Source Models: Open-source models refer to their source code being open to the public, allowing anyone to view, use, modify, and distribute the software model. In the fields of artificial intelligence and machine learning, open-source models have contributed to the rapid development and widespread application of technology. Through open-source, researchers and developers can share and improve models, accelerating the iteration and optimization of new technologies, while also making technology more transparent and accessible.
  • Parameter Count: In machine learning models, parameters are internal variables learned from training data, used for prediction or decision making. The number of parameters is directly related to the complexity and capability of the model. For instance, a model with a greater number of parameters may capture more complex patterns and relationships, but it may also require more data and computational resources for training. When comparing different models, the parameter count is an important performance indicator.

You May Also Like

This article is submitted by our user under the News Submission Rules and Guidelines. The cover photo is computer generated art for illustrative purposes only; not indicative of factual content. If you believe this article infringes upon copyright rights, please do not hesitate to report it by sending an email to us. Your vigilance and cooperation are invaluable in helping us maintain a respectful and legally compliant community.

Subscribe to our Newsletter

Get the latest in enterprise business and tech with exclusive peeks at our new offerings