Alibaba releases Qwen2.5-Max, a large mixture-of-experts model served via its cloud

Alibaba Qwen/Qwen models

Alibaba released Qwen2.5-Max, a large mixture-of-experts model pretrained on over 20 trillion tokens, available through Qwen Chat and Alibaba Cloud's API rather than open weights. Alibaba reported it beat DeepSeek V3 on benchmarks including Arena-Hard, LiveBench, LiveCodeBench and GPQA-Diamond.

• Large MoE model; pretrained on 20+ trillion tokens
• Accessible via Qwen Chat and Alibaba Cloud API; not open-weight
• Beats DeepSeek V3 on Arena-Hard, LiveBench, LiveCodeBench, GPQA-Diamond benchmarks
• Offered through cloud service; uses OpenAI-compatible API endpoints