China’s Alibaba takes another swipe at America’s AI supremacy

Aug 03, 2026 06:01 PM - 3 weeks ago 542

Chinese tech elephantine Alibaba released what it says is its largest and “most tin AI exemplary to date,” claiming capacity rivaling the champion systems from US frontier labs Anthropic and OpenAI, arsenic good arsenic home rivals for illustration Moonshot AI’s Kimi K3. Alibaba said it was making the model, Qwen3.8-Max, wide disposable to users successful a blog post published connected Monday. The merchandise had been expected aft the institution previewed the exemplary past month, erstwhile it claimed it was “second only to Fable 5,” Anthropic’s flagship.

The unfastened merchandise of different highly tin Chinese AI exemplary adds to sky-high tensions on aggregate fronts successful Silicon Valley and Washington complete really to safely negociate AI systems and clasp the US’ technological separator complete China.

Results from Alibaba’s ain testing shared connected Monday suggests it’s arsenic powerful arsenic claimed, arsenic does its ranking connected crowdsourced model-comparison level Arena.AI. Alibaba’s ain testing shows the model’s capacity to broadly lucifer — and sometimes transcend — that of Fable 5 connected benchmark tests. On the Arena matter exemplary leaderboard, Qwen3.8-Max trails only Fable 5 and 3 models successful Anthropic’s Opus family. For frontend coding, it is beaten only by 2 Claude Opus models and Kimi K3, and for visual analysis, only Fable 5 trounces it.

Alibaba says Qwen3.8-Max has 2.4 trillion parameters, a numerical measurement of the settings a exemplary learns during training that it uses to process data, admit patterns, and undertake various tasks. While parameter counts are wide utilized arsenic a shorthand for exemplary performance, a bigger number does not ever mean better. Moonshot’s Kimi K3 exemplary has 2.8 trillion parameters, but astir apical American labs support figures private, and neither OpenAI nor Anthropic disclose nonstop counts for their apical systems.

The Chinese institution said it will merchandise the weights for Qwen3.8-Max adjacent week. Weights are the adjustable numerical values of an AI that find really it processes information. Open-weight systems, while much restrictive than accepted open-source software, springiness developers acold much power than they person complete proprietary products from companies for illustration OpenAI and Anthropic. Qwen3.8-Max marks a return to open-weight releases for Alibaba aft the institution concisely pivoted towards proprietary releases for its much precocious models earlier this year.

Open-weight releases person go a growing constituent of differentiation for China’s AI industry, wherever they person go the norm. Moonshot released Kimi K3’s past week and galore different apical AI models are besides open-weight. Beijing has championed the strategy arsenic a intends of increasing China’s power successful world AI governance and encouraging wide take of home tech champions.

Alibaba’s merchandise intensifies title pinch China, whose firms look to beryllium quickly narrowing the spread pinch US companies and person accrued the tempo of releases successful caller weeks. Qwen3.8-Max intimately follows the merchandise of Kimi K3, viewed arsenic different situation to American AI dominance, and some ByteDance and MiniMax released tin caller video procreation models connected Friday.

Openness has besides proved to beryllium a divisive issue. Amid reports of imaginable crackdown connected unfastened devices successful the aftermath of the Chinese releases, the US manufacture has largely rallied astir preserving entree to open-weight models, some arsenic a information necessity and arsenic intends of preserving competition. That statement comes arsenic closed-model providers, notably OpenAI and Anthropic, look expanding scrutiny aft revealing a slew of cyberattacks unknowingly perpetrated by their ain escaped AI agents. Incident reports from 1 unfortunate propose the restrictive information rails intended to extremity nefarious usage of AI models besides limits their usage arsenic protect tools.

Follow topics and authors from this communicative to spot much for illustration this successful your personalized homepage provender and to person email updates.

More