What Alibaba released
Alibaba Group unveiled Qwen3.8-Max, described as the largest and most capable model in its Qwen series to date, with 2.4 trillion total parameters and a context window of up to 1 million tokens. The model is multimodal, handling both text and visual input, and is positioned for coding, research, workplace applications, and extended autonomous tasks. The Qwen3.8 API is available through the Qwen AI platform and has been integrated into Alibaba's new AI agent product, Qwen Office.
How Qwen3.8-Max compares
In recently released third-party Arena rankings, Alibaba's Qwen model ranked second only to Anthropic's Claude series. Alibaba claims the model offers competitive performance while costing less to use than several rival flagship AI models. By total parameter count, Qwen3.8-Max trails domestic rival Moonshot AI's recently launched Kimi K3, which carries 2.8 trillion parameters, though Alibaba emphasized that higher parameter counts do not automatically translate into better model behavior.
QwenWork: the new enterprise agent
On the same day as the model launch, Alibaba opened public testing of QwenWork, an enterprise-level AI agent created by combining three previously separate workplace-agent products: QoderWork, MuleRun and Wukong. Alibaba framed the consolidation as part of a broader push to compete both in frontier AI models and in AI-powered office software, centralizing its agent offerings under a single enterprise platform.
Competitive context in China
The launch came on the same day that Chinese AI research firm DeepSeek introduced its V4-Flash model, highlighting the rapid pace of development among Chinese AI labs and the growing popularity of open-weight AI models. Coverage noted that China's AI companies appear to be narrowing the gap with their US rivals.
Open-source plans
Alibaba is expected to open-source Qwen3.8-Max the following week, alongside the Qwen3.8-27B model, extending the release beyond the proprietary API and integrated Qwen Office deployment.
What remains to be seen
Independent benchmarks, real-world enterprise adoption of QwenWork, and uptake of the open-sourced Qwen3.8-27B release will clarify how the new model and consolidated agent platform perform against existing offerings from both Chinese and US competitors.
Share this article
