Alibaba has introduced Qwen3.8, a new multimodal AI model, claiming it is currently second only to Anthropic's Fable 5. The company unveiled the model at the World Artificial Intelligence Conference in Shanghai on July 20, 2026. Qwen3.8 contains 2.4 trillion parameters, marking the first time the Qwen family has reached over a trillion parameters while maintaining multimodal capabilities, including the processing of images, video, and documents. Alibaba reports that the model is specifically optimized for coding tasks, full-stack development, and data analysis. Following the announcement, Alibaba shares rose by as much as 5.4% on Monday.

The model is currently accessible through Alibaba’s Token Plan subscription and Qoder developer tools, with a trial period priced at 10% of the standard rate. Although the company intends to release the model weights publicly, it has not provided a specific date or licensing details. This launch follows a competitive move by Moonshot, which recently released the 2.8-trillion-parameter Kimi K3 model.

Despite the performance claims, Alibaba has yet to provide benchmark scores or an independent model card to verify its ranking. Previous flagship releases, such as Qwen3.7-Max, included published test results, a practice not continued here. Currently, no public leaderboards have evaluated Qwen3.8, leaving researchers unable to validate the team's assertions until the weights are released. Analysts view this release as part of an escalating competition among Chinese labs to provide large-scale, open-weight models that rival the capabilities of frontier systems from US-based laboratories.

Alibaba has introduced Qwen3.8, a new multimodal AI model, claiming it is currently second only to Anthropic's Fable 5. The company unveiled the model at the World Artificial Intelligence Conference in Shanghai on July 20, 2026. Qwen3.8 contains 2.4 trillion parameters, marking the first time the Qwen family has reached over a trillion parameters while maintaining multimodal capabilities, including the processing of images, video, and documents. Alibaba reports that the model is specifically optimized for coding tasks, full-stack development, and data analysis. Following the announcement, Alibaba shares rose by as much as 5.4% on Monday.

The model is currently accessible through Alibaba’s Token Plan subscription and Qoder developer tools, with a trial period priced at 10% of the standard rate. Although the company intends to release the model weights publicly, it has not provided a specific date or licensing details. This launch follows a competitive move by Moonshot, which recently released the 2.8-trillion-parameter Kimi K3 model.

Despite the performance claims, Alibaba has yet to provide benchmark scores or an independent model card to verify its ranking. Previous flagship releases, such as Qwen3.7-Max, included published test results, a practice not continued here. Currently, no public leaderboards have evaluated Qwen3.8, leaving researchers unable to validate the team's assertions until the weights are released. Analysts view this release as part of an escalating competition among Chinese labs to provide large-scale, open-weight models that rival the capabilities of frontier systems from US-based laboratories.