Alibaba previews 2.4 trillion-parameter Qwen3.8 Max, claims it trails only Fable 5

Alibaba Group Holding Ltd. released a preview version of its Qwen3.8 Max model on July 19, 2026, claiming the 2.4 trillion-parameter multimodal system rivals frontier AI models and outperforms all but Anthropic's Fable 5. The company said it plans to make the model open-weight soon.

The Qwen team posted on X that it believes Qwen3.8 Max is one of the most powerful models available today, comparable to leading proprietary AI systems and second only to Fable 5. The preview is accessible through Alibaba's Token Plan, Qoder, and QoderWork platforms.

Technical specifications and positioning

Qwen3.8 Max is Alibaba's first trillion-parameter multimodal model. Its 2.4 trillion parameters place it among the largest openly described AI systems to date, exceeding the size of many frontier models. Shuai Bai, a member of the Qwen team, said on X that the model delivers multimodal understanding competitive with and often ahead of current proprietary models.

• Parameter count: 2.4 trillion, making it Alibaba's largest AI model by a wide margin
• Multimodal capability: processes text, images, and other data types
• Competitive claims: Alibaba says it beats GPT-5.6 on some benchmarks and trails only Fable 5
• Release cadence: preview launched just days after Moonshot's open-weight Kimi K3 release

Implications for the AI landscape

The release of Qwen3.8 Max preview continues a pattern of rapid iteration from Chinese AI labs. Moonshot released its open-weight Kimi K3 model days earlier, and GLM is expected to announce a new model soon. The gap between US proprietary models and Chinese open-weight systems continues to narrow, according to AI researchers monitoring the space.

Alibaba's commitment to open-weight distribution could pressure other large model developers to follow suit. The company has not specified a definitive date for the open-weight release but described it as coming soon. Once released, researchers and developers will be able to inspect, modify, and self-host the model, a contrast to the API-only access of many rival systems.

For now, the preview version lets users test Qwen3.8 Max through Alibaba's paid products. The company has not published independent benchmark results or third-party evaluations. The model's actual performance relative to Fable 5 and GPT-5.6 will be clearer once the open-weight version is released and subjected to wider scrutiny.