How many people work on ByteDance's AI model development?

Seed, ByteDance's model development team led by former Google DeepMind scientist Wu Yonghui, has about 2,000 members in China and overseas, including core researchers, infrastructure engineers, a data labeling team, and translators.

What is model distillation and why is ByteDance avoiding it?

Model distillation is the process of compressing a large, complex AI model into a smaller, faster one by training the smaller model to copy the knowledge and outputs of the larger teacher. ByteDance has implemented an independent approach that does not involve distillation for more than a year, which some believe has led to slower development versus rivals, but management believes only independent development can result in a model that outperforms rivals.

How popular is ByteDance's consumer AI model?

Doubao, ByteDance's flagship consumer-facing model, is the most popular in China with 324 million monthly active users.

Large Language ModelsAI Business & IndustryArs Technica AIAug 7, 2026Aug 8, 2026, 01:01 JST

ByteDance builds AI team to rival Anthropic, spurns model distillation

Send on LINE

3 Key Points

What happened
ByteDance's Seed model development team, led by former Google DeepMind scientist Wu Yonghui and comprising about 2,000 members across China and overseas, is pursuing independent AI model development without relying on distillation (compressing knowledge from other labs' models). Founder Zhang Yiming told the team two weeks ago to target "world-leading model capabilities" in the long run without worrying about near-term competitive gaps.
Why it matters
ByteDance has invested in AI more aggressively than any other Chinese tech giant over the past three years, building data centers and expanding its cloud unit Volcano Engine. Its consumer model Doubao is the most popular in China with 324 million monthly active users, and its SeeDance video-generation model ranks among the most advanced globally. This independent approach suggests ByteDance believes only developing models from scratch—rather than adapting rivals' work—can produce genuinely superior systems.
What to watch
ByteDance's management acknowledges this strategy may cause slower near-term development versus competitors but is betting on long-term capability leadership. The company also has ambitions to develop custom AI chips, which could further reduce reliance on external technology.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

ByteDance's approach to AI development diverges sharply from the prevailing industry practice of model distillation, a cost-effective shortcut that has become standard among competitors. By rejecting this faster path, ByteDance is betting that investing heavily in independent research and infrastructure will ultimately yield models with genuinely superior capabilities. This strategy aligns with founder Zhang Yiming's recent directive to the Seed team, which acknowledged the near-term cost of this approach but framed it as necessary for long-term leadership.

The company's three-year investment spree—outpacing other Chinese tech giants in AI spending—has already produced commercially successful models: Doubao's 324 million monthly active users in China represent a significant consumer foothold, and SeeDance's ranking among the most advanced video-generation models globally suggests the independent approach is yielding results. ByteDance's plans to develop custom AI chips further underscore its commitment to reducing dependency on external technology and building proprietary competitive advantages.

FAQ

How many people work on ByteDance's AI model development?: Seed, ByteDance's model development team led by former Google DeepMind scientist Wu Yonghui, has about 2,000 members in China and overseas, including core researchers, infrastructure engineers, a data labeling team, and translators.
What is model distillation and why is ByteDance avoiding it?: Model distillation is the process of compressing a large, complex AI model into a smaller, faster one by training the smaller model to copy the knowledge and outputs of the larger teacher. ByteDance has implemented an independent approach that does not involve distillation for more than a year, which some believe has led to slower development versus rivals, but management believes only independent development can result in a model that outperforms rivals.
How popular is ByteDance's consumer AI model?: Doubao, ByteDance's flagship consumer-facing model, is the most popular in China with 324 million monthly active users.

Ars Technica AIRead Original Article

Send on LINE

Get the latest Large Language Models news every morning

For example, today's edition would include:

Agentic AI pushes identity to front of data security, Oracle saysSiliconANGLE AI · 20m ago
Qiagen grounds drug discovery agents in 25+ years of curated dataSiliconANGLE AI · 20m ago
Momentic launches Mo, an AI agent that tests apps without scriptsSiliconANGLE AI · 20m ago

AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.

Subscribe free →

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.