Alibaba Challenges America’s Dominance in AI Technology
Chinese tech powerhouse Alibaba has unveiled what it claims is its largest and most advanced AI model yet, asserting that its performance rivals the top systems from leading US labs like Anthropic and OpenAI, as well as domestic competitors such as Moonshot AI’s Kimi K3. In a blog post released on Monday, Alibaba announced that the model, named Qwen3.8-Max, would be made widely accessible to users. This launch was anticipated after the company teased the model last month, stating it was “second only to Fable 5,” Anthropic’s flagship offering.
The public release of this highly capable AI model adds to the already heightened tensions in Silicon Valley and Washington regarding the safe management of AI systems and the need to maintain the US’s technological edge over China.
Results from Alibaba’s internal testing, shared on Monday, indicate that the model performs as claimed, supported by its ranking on the crowdsourced model-comparison platform Arena.AI. According to Alibaba’s testing, Qwen3.8-Max’s performance generally matches or even surpasses that of Fable 5 in benchmark tests. On the Arena text model leaderboard, it ranks just behind Fable 5 and three models from Anthropic’s Opus family. In frontend coding tasks, it is outperformed only by two Claude Opus models and Kimi K3, while in visual analysis, Fable 5 remains the only model that significantly outshines it.
Alibaba states that Qwen3.8-Max boasts 2.4 trillion parameters, which are the adjustable settings a model learns during training to process data, recognize patterns, and perform various tasks. While parameter counts are often used as a shorthand for assessing model performance, a higher number doesn’t always equate to better performance. For instance, Moonshot’s Kimi K3 has 2.8 trillion parameters, but many leading American labs keep their figures under wraps, and neither OpenAI nor Anthropic disclose the exact counts for their top models.
The company plans to release the weights for Qwen3.8-Max next week. Weights are the adjustable numerical values that dictate how an AI processes information. Open-weight systems, while more restrictive than traditional open-source software, provide developers with significantly more control compared to proprietary products from companies like OpenAI and Anthropic. This marks a return to open-weight releases for Alibaba after a brief shift towards proprietary models for its more advanced offerings earlier this year.
Open-weight releases have become a key differentiator for China’s AI sector, where they are increasingly common. Moonshot recently released Kimi K3, and many other leading AI models are also open-weight. Beijing has promoted this strategy as a way to enhance China’s influence in global AI governance and to foster the widespread adoption of domestic tech leaders.
Alibaba’s release heightens the competition with China, where firms are rapidly closing the gap with US companies and have accelerated their release schedules in recent weeks. Qwen3.8-Max follows closely on the heels of Kimi K3, which is seen as another challenge to American AI supremacy, while both ByteDance and MiniMax launched new video generation models just last Friday.
The issue of openness has also sparked debate. Amid reports of potential crackdowns on open tools following these Chinese releases, the US industry has largely united around the need to preserve access to open-weight models, viewing it as essential for safety and competition. This discussion arises as closed-model providers, particularly OpenAI and Anthropic, face increasing scrutiny after disclosing a series of cyberattacks inadvertently caused by their own AI agents. Incident reports from one victim suggest that the restrictive safety measures intended to prevent malicious use of AI models may also hinder their effectiveness as defensive tools.
