Alibaba has introduced Qwen2.5-Max, a large-scale Mixture-of-Expert (MoE) AI model, pretrained on over 20 trillion tokens and fine-tuned using advanced techniques like Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF).
Qwen2.5-Max outperforms leading models like DeepSeek V3 in benchmarks such as coding (LiveCodeBench) and human-like preferences (Arena-Hard). The model is now accessible via Alibaba Cloud’s API, compatible with OpenAI API standards, and integrated into Qwen Chat for interactive use.
This launch highlights Alibaba’s focus on scaling AI intelligence, with ongoing efforts to enhance reasoning and unlock unprecedented capabilities in future iterations.
