Alibaba has launched Qwen3.8-Max, its largest artificial intelligence model to date, as competition among Chinese AI developers continues to intensify. The new multimodal model combines 2.4 trillion parameters with a Mixture-of-Experts (MoE) architecture designed to deliver high performance while reducing computing costs.
Unlike traditional AI models that activate every parameter for each request, Qwen3.8-Max uses roughly 95 billion active parameters during inference. Alibaba said the approach improves efficiency by activating only the parts of the model needed for each task, helping reduce latency and computing demands.
Built for Large-Scale AI Workloads
Qwen3.8-Max processes text, images and video. It also supports a context window of up to one million tokens, enabling it to analyze lengthy documents, extended conversations and complex datasets.
Alibaba also demonstrated the model completing a software engineering project over 16 days, showcasing its ability to sustain long-running coding tasks.
Competition in China’s AI Market
The launch comes as Chinese AI developers accelerate efforts to compete on model performance, efficiency and deployment costs. Qwen3.8-Max joins a growing field of advanced foundation models that includes Moonshot AI’s Kimi K3 and DeepSeek’s V4-Flash.
While Kimi K3 features 2.8 trillion parameters and DeepSeek V4-Flash uses a smaller 284-billion-parameter architecture, each model takes a different approach to balancing capability, efficiency and operating costs.
Alibaba said Qwen3.8-Max is priced to compete with other leading AI models, while DeepSeek’s V4-Flash has attracted attention for its low-cost inference pricing. The emphasis on affordability shows that Chinese AI developers are now competing on efficiency as much as model performance.
Shortly after its release, Qwen3.8-Max ranked as the highest-rated Chinese text model on the Arena.AI leaderboard. The model also placed among the leading multimodal systems on the platform, reflecting Alibaba’s continued progress in language and visual reasoning tasks.
Open-Weight Strategy Expands Developer Choice
Beyond benchmark rankings, the launch reflects a broader shift across China’s AI industry. Companies are no longer competing on model size alone. They are also racing to improve efficiency, lower deployment costs and give developers more flexibility.
Meanwhile, Alibaba plans to release Qwen3.8-Max as an open-weight model, continuing a strategy that provides developers with access to the model weights alongside hosted API services. The approach allows organizations to deploy the model on their own infrastructure or through third-party providers, unlike many leading Western AI systems that remain closed.
The release of Qwen3.8-Max underscores the pace of China’s AI development as domestic companies continue to challenge global rivals such as OpenAI, Anthropic and Google. As the AI race evolves, success will depend not only on building larger models but also on delivering systems that are efficient, affordable and practical for real-world deployment.
