This Alibaba AI model is built on the Mixture-of-Experts (MoE) architecture and has been trained on over 20 trillion tokens. According to the developers, Qwen2.5-Max has demonstrated “significant progress in intelligence capabilities” and is now ready for use.
Initial results show that Qwen2.5-Max is more powerful than its competitors.
Qwen2.5-Max not only improves performance and accuracy, but is also better able to handle tasks that require deep understanding of context, such as text analysis, translation, and content creation. "Qwen2.5-Max demonstrates significant progress in handling complex queries and providing relevant answers," Alibaba said in a blog post.
Impressive numbers from Alibaba's new model
The Qwen2.5-Max model has been tested on many important benchmarks, including MMLU-Pro, LiveCodeBench, LiveBench, and Arena-Hard. The results show that Qwen2.5-Max outperforms DeepSeek V3 in tests such as Arena-Hard, LiveBench, and LiveCodeBench, and shows competitive results in MMLU-Pro. Compared with other leading models such as GPT-4o and Claude-3.5-Sonnet, Qwen2.5-Max also asserts its leading position.
Alibaba Cloud plans to integrate Qwen2.5-Max into its cloud services to help customers solve a variety of problems, from automating data processing to improving customer interactions through chatbots and optimizing business processes. The AI model is now available through the Qwen Chat service, allowing users to interact, test its capabilities and experiment with various functions. In addition, the API is also open to developers. To access it, users need to register with Alibaba Cloud, activate the Model Studio service and create an API key.
Source: https://thanhnien.vn/alibaba-ra-mat-mo-hinh-ngon-ngu-canh-tranh-voi-gpt-4o-va-deepseek-v3-185250129152256505.htm
Comment (0)