ByteDance: Are Chinese AI Models Catching Up to the US? | AI Magazine
Article
Artificial Intelligence (AI)
ByteDance: Are Chinese AI Models Catching Up to the US?
By Adam Pond<br>August 10, 2026
undefined mins
Share this article
Prioritise Us on Google
Share this article
Prioritise Us on Google
The US and China are increasingly locked into an AI race, with the US slightly ahead on frontier models. Credit: Getty Images
ByteDance reportedly has a new model in development with an impressive parameter count as Chinese labs seek to close the gap with US firms
ByteDance is training an AI model that could approach the size of Anthropic’s most cutting-edge Mythos system, the Financial Times reports.<br>The massive Chinese technology firm, which owns the popular social media platform TikTok, is aggressively expanding into AI, including by building out data centres.<br>It follows Chinese AI startup Moonshot releasing Kimi K3 last month, a 2.8 trillion parameter model built with native vision capabilities and a 1-million-token context window.
ByteDance’s model in development<br>The FT spoke with three people with knowledge of the matter who said the massive Chinese tech firm ByteDance is in the early stages of training a model with as many as 10 trillion parameters.<br>Model parameters as the internal configuration variables of a machine learning model that control how it processes data and makes predictions.<br>While one of the US market leaders in AI, Anthropic, doesn’t disclose the size of its models, industry estimates cited by the FT say its most advanced Mythos 5 has about 8 trillion parameters and Fable 5 about 5 trillion.<br>The news follows Chinese firms Moonshot and Alibaba releasing AI models that are increasingly seen as competitors to widely used US models.<br>Closing the gap<br>According to information released by Moonshot, its Kimi K3 model outperformed OpenAI’s GPT 5.5 and Anthropic’s Opus 4.8 across most benchmarks, including coding, general agent and visual agent tasks.<br>Alibaba used the platform Arena.ai to compare its latest offering, Qwen3.8-Max, to other models. It ranked fifth in Text Arena and second in Vision Arena, with only Anthropic’s Claude models ahead of it in the areas.<br>Alibaba’s new model boasts an impressive 2.4 trillion parameters, slightly trailing behind MoonShot’s recent 2.8 trillion parameter model.<br>The reported new offering from ByteDance looks to not only top Chinese models in parameters, but also close the gap with some of the US frontier models.
President Xi and President Trump (Credit: Getty)
The AI race<br>The news of ByteDance’s new model in development is the latest move in the race between AI firms in the US and China, which are increasingly locked into an AI race, with the US slightly ahead on frontier models.<br>However, Chinese AI firms are increasingly catching up – all while US AI firms like Anthropic have accused Chinese counterparts of illicitly extracting capabilities through a process called distillation.<br>It is believed by many that the ‘race’ dynamics only add to the potential risks for the rapidly developing technology. The Center for AI Safety has argued that nations and corporations are competing to rapidly build and deploy AI in order to maintain power and influence.<br>It argues this is similar to the nuclear arms race during the Cold War and participation in the AI race may serve individual short-term interests, but ultimately amplifies global risk for humanity.<br>Distillation accusations against Chinese firms<br>This news builds on reports of distillation by leading Chinese firms. In February 2026, Anthropic accused Chinese AI labs DeepSeek, Moonshot and MiniMax of “industrial-scale campaigns” to “illicitly extract Claude’s capabilities to improve their own models”.<br>The leading US AI firm said the attacks were growing in intensity and sophistication and stressed it required rapid, coordinated action among industry players.<br>Dario Amodei, Anthropic's CEO, previously said: "Distillation does not allow the Chinese Communist Party to obtain equivalent or superior AI capabilities to the US, but it can bring the Chinese frontier to within a few months of the US frontier."
CEO Dario Amodei is pictured during a 2023 Senate Judiciary Committee hearing. Picture: Getty Images
However, ByteDance is reportedly against using techniques like this to rapidly advance the technology.<br>On 6 August, Chinese state-backed newspaper The Paper learned from sources within ByteDance that its founder Zhang Yiming has stated that model development should adhere to long-termism and delayed gratification, rather than using others' output to gain temporary rankings.<br>It is understood by the Chinese paper that ByteDance strictly prohibits distillation of open-source models and has even strengthened related restrictions internally through API testing and other means.
Company portals
Alibaba Group
Anthropic
ByteDance
Moonshot
OpenAI
Executives
Dario Amodei<br>CEO and Co-Founder
Yiming Zhang<br>Founder