Alibaba Cloud has expanded its Qwen family with a new flagship model named Qwen3.8 Max Preview. Unveiled at the World Artificial Intelligence Conference in Shanghai, the massive model boasts 2.4 trillion parameters and claims performance second only to Anthropic Claude Fable 5. But the company has yet to release benchmark details or a direct comparison with its previous model.
Alibaba has kept the technical architecture and the exact number of active parameters secret. This makes it impossible to verify the claims of near top tier performance. It also leaves developers without a clear picture of how this new release improves upon Qwen3.7 Max, which launched earlier this year. Without transparent data, the industry is left to take the performance assertions at face value.
Access to the preview is currently restricted to Token Plan, Qoder, and QoderWork platforms. During this initial testing phase, query pricing is set at just 10% of the standard rate. Alibaba intends to release the weights for the model soon, a promise that was never fulfilled for the prior Qwen3.7 Max version. Open sourcing these weights would mark a major shift in how the company shares its foundational models with the wider developer community.
