Alibaba Launches Qwen3.8-Max and DeepSeek Cuts Inference Costs: China's AI Race Heads Toward Cheaper Models

Qwen3.8-Max is Alibaba's latest and largest language model, using a mixture-of-experts (MoE) architecture with 2.4 trillion total parameters, but activating only about 95 billion per request, reducing compute costs.

DeepSeek V4-Flash offers inference pricing lower than several competing systems, making it an attractive option for companies seeking strong performance at lower operational cost.

MoE allows running massive models with higher efficiency and lower cost, enabling organizations in the region to deploy advanced AI applications without heavy infrastructure investment.

Yes, especially with falling costs and improved performance; teams can pilot these models for specific use cases while considering local privacy and compliance requirements.