Industry reports and surveys from credible tech industry analysts

Alibaba's Qwen2.5-Max AI Model Claims to Surpass DeepSeek-V3, GPT-4o, and Llama-3.1-405B

Jan 29, 2025, 10:13 AM

Alibaba Group has unveiled its new AI model, Qwen2.5-Max, which the company claims outperforms DeepSeek-V3 and other leading models. The model utilizes a large-scale Mixture-of-Experts (MoE) architecture and is pretrained on massive data, followed by fine-tuning with curated Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF) techniques. According to announcements on Alibaba Cloud's official WeChat account, Qwen2.5-Max surpasses GPT-4o, DeepSeek-V3, Llama-3.1-405B, and is comparable to Anthropic's Claude 3.5 Sonnet across various benchmarks, including Arena Hard, LiveBench, LiveCodeBench, and GPQA-Diamond. Alibaba anticipates that the next version of Qwen will be even stronger due to improved post-training methods.