ByteDance founder Zhang Yiming has said the company will not rely on AI distillation to advance its models, even if that means falling behind competitors, according to The Information. People familiar with the matter said the decision is partly influenced by ByteDance’s complex relationship with the U.S. government over its ownership of TikTok.
In this context, distillation refers to training an AI model using the outputs of a more advanced frontier model, typically one developed by another company. Anthropic has accused several Chinese AI firms—including DeepSeek, Moonshot, MiniMax, Z.ai and Alibaba—of distilling its Claude models. ByteDance, however, has not been accused of doing so.
READ: China’s ByteDance partners with Samsung to develop AI chip (February 11, 2026)
Analysts note that a no-distillation pledge is difficult to verify from outside the company. Neither Zhang nor ByteDance spell out how the company plans to enforce the policy internally, whether it applies to synthetic data generated by its own smaller models, or which performance benchmarks Zhang is willing to sacrifice in the near term.
Zhang’s remarks have been seen as notable since he is known for keeping a low profile in recent years. However, South China Morning Post recently mentioned the founder “has been sitting in on the Seed team’s core technology reviews and reading OpenAI papers late into the night.”
This comes during a happening time in the Chinese AI industry. Alibaba recently unveiled its largest and most capable AI model to date, Qwen3.8-Max, sending its shares higher.
Meanwhile, a research firm said DeepSeek’s latest model offers pricing that is more than 100 times cheaper than Anthropic’s Claude Fable 5, underscoring the intensifying race to deliver powerful AI at lower cost.
READ: TikTok owner ByteDance takes on Alibaba by entering China’s AI cloud market (January 20, 2026)
Another AI model that made waves was Moonshot’s Kimi K3, a 2.8-trillion-parameter open-weight model designed for software engineering and autonomous task execution. Michael Kratsios, who serves as director of the White House Office of Science and Technology Policy, has alleged that Moonshot AI distilled Anthropic’s Fable model when building Kimi K3, its AI system. He stated in an X post that Moonshot ran the operation via a purpose-built internal platform, which rotated access routes in a bid to evade detection and used NVIDIA GB300 servers in Thailand, which are barred from being exported to China by U.S. export rules.
Earlier this year, ByteDance rolled out its new audio and video model, Dreamina Seedance 2.0, on its editing platform CapCut. The company said the model allows creators to draft, edit, and sync video and audio content by using prompts, images, or reference videos.