Alibaba’s Qwen 3.8 Max preview outperforms competitors in speed, signaling a new AI arms race with crypto implications

Alibaba just dropped what might be the most capable AI model to come out of China yet. Qwen3.8-Max, previewed at the World AI Conference in Shanghai on July 19, packs 2.4 trillion parameters and has been clocking faster execution times than most of the competition in real-world coding tests. What Qwen 3.8 Max actually does differently The model runs on a sparse Mixture-of-Experts (MoE) architecture. Instead of firing up all 2.4 trillion parameters for every query, it selectively activates only the relevant “expert” sub-networks for each task. This design lets Qwen3.8-Max handle text, images, video, and documents while staying computationally efficient. It’s the first trillion-plus parameter multimodal model from Alibaba’s Qwen series. Independent user assessments found the model generating and rendering Three.js 3D scenes faster than several top competitors. In those tests, Qwen3.8-Max reportedly placed second only to Anthropic’s Claude Fable 5 in overall performance. It outperformed both Moonshot AI’s Kimi K3 and OpenAI’s GPT-5.6 Sol in coding tasks specifically. The model also features a context window of approximately 1 million tokens. Alibaba is making the model available throu...