Alibaba Qwen 3.8 Max Shows China Closing in on U.S. Models

Robert Way via Getty ImagesChinese tech giant Alibaba’s new AI model shows that enterprises will likely continue to be attracted to Chinese AI models that are not only nearly as capable as but also far cheaper than frontier models from U.S. providers.Alibaba previewed its new flagship model Qwen 3.8 Max on July 19. The model boasts 2.4 trillion parameters, and the Chinese vendor claims that it ranks just behind Anthropic’s Fable 5 in performance. While the model is currently available to test on Alibaba’s Qoder platform, the vendor said it will release the weights soon.Qwen 3.8 Max was introduced a few days after Chinese AI startup Moonshot AI launched Kimi K3, its open-weight model with 2.8 trillion parameters.Meanwhile, on Tuesday, Moonshot said it was experiencing intense demand for Kimi K3 -- considered the world’s largest open model -- and was struggling to provide enough compute capacity to meet it, the AP and others reported. Moonshot said it has paused individual subscriptions until it can provide more capacity.Related:Startup Launches Foundry to Develop Materials for ChipsAs for Qwen, the Alibaba model is designed for advanced coding and agentic tasks, capabilities sought by enterprises. Its release reflects the booming model competition in the Chinese AI market.A Competitive Choice“It’s a well-timed response,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget, referring to the release of the model. “Alibaba Cloud doesn’t want to fall behind its domestic competitors.”While Alibaba is one of the top AI vendors in China, big tech vendors such as Baidu and Tencent are legitimate rivals in the fast-growing Chinese AI market.“You want companies to be competitive,” Su said. “You want vendors to be up against each other because competition brings only benefit to the end user.”The U.S. and ChinaIt is now clear that Chinese AI vendors are producing models that are compelling alternatives to those from Anthropic, OpenAI and Google.“It’s realistic to think that within a year, there will be no question that China will have models that will be just as powerful as whatever is available now in the U.S.,” said David Nicholson, an analyst at Futurum Group.He added that, in the near term, discussions about model choice will focus on the cost of models for achieving certain tasks and on how models are evaluated for factors such as security, governance and compliance.But amid geopolitical tensions between the U.S. and China, the U.S. government may be considering how enterprise access to Chinese models could affect national security.“Those [enterprise choices] will be affected by what the government decides about things like having Chinese models be at the center of fill-in-the-blank operations of the U.S.,” Nicholson said.Related:As AI Spending Climbs, Enterprises Get Serious About Token CostsThe Enterprise FactorNevertheless, it is also unlikely that U.S. enterprises will use the new models from Alibaba or Moonshot right away. The surge in demand for Kimi K3 is likely driven by individual developers, not organizations seeking enterprise subscriptions, according to the analysts.“No existing legacy company is going to be implementing anything like this into their Oracle or SAP or Microsoft-based worlds at this point, or Google-based worlds at this point,” Nicholson said, adding that this is because of the tensions between China and the U.S “It's still a way out.”Su said that U.S. enterprises tend to be more careful with Chinese models.“The concerns around data residency play a big role,” he said. “It’s less about geopolitics or regional biases.”As to Alibaba’s claim that Qwen 3.8 Max is just a little behind Fable 5, Nicholson said there is no way to verify it since Alibaba did not use traditional benchmarks, but independent developer benchmarks that are harder to verify. The claim could be a competitive tactic, he said.“There is a massive incentive if you can crush your opponents by merely making a claim,” he said. “We don’t know yet what the reality is.”Related:OpenAI Unveils GPT-Red to Test AI Model SafetyHe added that even if the models rival those of frontier model providers in the U.S., it might be more advisable for enterprises to stick with their existing frameworks.“Trust Salesforce and Oracle and ServiceNow and Google and Microsoft to vet the way that models are used, the way that agentic AI is deployed, because they have better control over context and the governance and compliance frameworks that already exist,” he said.Despite these challenges, there is no doubt that Chinese models offer a more cost-effective option. While Alibaba plans to make Qwen 3.8 Max open-weight, pricing remains unclear. Kimi K3 costs $3 per million input tokens and $15 per million output tokens. Claude Fable is $10 per million input tokens and $50 per million output tokens through Anthropic Claude.About the AuthorNews Writer, AI BusinessEsther Shittu has covered AI technologies and industry trends since 2021. As co-host of the Targeting AI podcast, she talks with experts, thought leaders and practitioners exploring critical AI developments. Before AI Business, she wrote for SearchEnterpriseAI, the New York Daily News, Bklyner and the Brooklyn Daily Eagle. When she's not diving deep into the world of AI, she spends her time on passion projects and raising her three daughters.