ByteDance's own founder is putting a multi-year horizon on closing the model gap, so its near-term AI bets rest on agents, video and spending on infrastructure.
ByteDance plans to release an AI assistant modeled on Meta’s Muse within weeks, and co-founder Zhang Yiming has told the company’s AI team that ByteDance can compete with OpenAI and Anthropic at the cutting edge, but not before 2029 or 2030. Bloomberg reported the plans on November 11, citing people familiar with an all-hands meeting of the AI team, according to The Next Web and IT之家. Engadget put Zhang’s estimate as "until at least 2029 or 2030" for ByteDance’s AI work to catch up.
The meeting also produced an unusually candid assessment of ByteDance’s model unit, Seed. Zhang reportedly told staff that the datasets used to pretrain Seed’s language models are orders of magnitude smaller than those of the leading US labs. IT之家 described that remark as previously unreported. Bloomberg’s sources did not say what the new assistant will be called or how it will work, and ByteDance did not respond to Bloomberg’s request for comment.
Zhang, who is 43, now spends much of his time behind the scenes shaping ByteDance’s AI strategy, IT之家 noted, adding that ByteDance was an early mover in Chinese AI before DeepSeek and others pulled ahead. His plan, as relayed by people familiar with the meeting, pairs a fast product launch and the Seedance video models with a frank, multi-year timeline for the core language models.
Zhang sets out an agent, a budget and a video target
Muse, which Meta introduced in September, is a voice-based agent with an animated avatar that runs on Meta’s smart glasses, and Meta is preparing a portable device for it called the Muse Charm, The Next Web reported. Engadget said Muse has more than 10 million users. Bloomberg’s sources said it is unclear whether ByteDance will build hardware of its own. According to IT之家, ByteDance hopes the push will help it narrow the gap with top US labs in about three years.
Zhang’s core plan, as IT之家 summarized it, includes spending up to US$70 billion and turning Seedance into China’s leading video AI platform. The Next Web described the sum as spending on AI infrastructure to strengthen ByteDance’s lead in video. The figure matches an earlier Bloomberg report, published in Chinese by Yahoo Finance Hong Kong on May 27, that ByteDance was discussing capital expenditure of up to US$70 billion in 2026, mostly on AI infrastructure such as data centers. That report put 2025 capital spending at about US$25 billion, said a large part of the new budget would come from about US$50 billion of 2025 profit, and called the plan preliminary and subject to change.
Seedance is where ByteDance already leads. The Next Web said the model family filled the gap OpenAI left when it shut down Sora, and Engadget noted that Seedance clips went viral earlier this year and drew legal scrutiny from US studios.
The data gap explains the long timeline
Zhang’s comment on pretraining data goes to the heart of how far Seed has fallen behind. Seed has not released a language model near the top of global benchmarks and has not opened its weights to developers, according to The Next Web and IT之家. Its last major language-model update, Seed 2.1, came in June.
Zhang also repeated that ByteDance should not rely on distillation, the practice of training a smaller model on a larger model’s outputs, and should take a longer view, IT之家 reported. The Next Web noted that Alibaba, Moonshot AI and Z.ai have been criticized over distillation by White House officials and by Anthropic chief executive Dario Amodei. Refusing that shortcut while working from less pretraining data leaves ByteDance to close the gap mainly by gathering more data and computing power, which helps explain both the 2029 to 2030 horizon and the scale of the spending plan.
Seed has roughly 1,000 researchers and engineers and is led by Wu Yonghui, whom Zhang hired from Google DeepMind in 2025. ByteDance gives Seed staff a preferential equity incentive scheme separate from the rest of the company, and Seed is also working on AI for robotics, virtual reality and other hardware, IT之家 reported. Wu oversaw a reorganization of Seed into four first-level departments in August.
An assistant launch within weeks is the first marker
The quickest test of Zhang’s plan is the assistant itself. If ByteDance releases it in the coming weeks, its features will show how closely it follows Muse’s voice-and-avatar design and how it relates to Doubao, which The Next Web called China’s most popular ChatGPT-style chatbot. Any hardware companion would widen the bet considerably.
Two further signals will show how much of the reported timeline is already moving. One is the next major Seed language model after Seed 2.1, and whether it narrows the benchmark gap Zhang acknowledged. The other is how Seed’s new safety and alignment unit grows. ByteDance created that unit in October, which The Next Web described as a first among major Chinese labs, and IT之家 said it hired about 10 core researchers for it. A capital spending figure confirmed by ByteDance itself would put a firm number on the plan.