ByteDance will not use distillation to speed up its AI models even if that leaves it behind Chinese rivals, founder Zhang Yiming told an all-hands for the company’s Seed AI team last month, according to Juro Osawa and Qianer Liu at The Information. Two employees present said the reasoning traces back to TikTok, since training on the outputs of American frontier models would invite a fresh round of Washington scrutiny, and ByteDance still runs TikTok’s commercial business worldwide after selling most of the US data security operation last year. The restraint has left ByteDance’s language models trailing Alibaba, DeepSeek, Z.ai, and Moonshot, and its newest one, Seed 2.1 Pro, doesn’t appear on the Artificial Analysis leaderboard at all.
