Zhang Yiming says ByteDance may not catch up with OpenAI until 2029

On November 12, 2026, The Paper reported, citing multiple informed sources, that ByteDance founder Zhang Yiming had again discussed the company's AI business in an internal sharing session, holding an all-hands meeting with the AI team. At that meeting, Zhang said ByteDance is capable of competing with OpenAI and Anthropic, and that its large models aim to match the international first tier. He added, however, that truly closing the gap with frontier technology would take until at least 2029 or 2030.

Zhang also acknowledged that, compared with leading overseas AI companies, the pre-training dataset behind ByteDance's Seed model series remains smaller, and called this the issue ByteDance AI needs to focus on most right now. A separate foreign media report cited in the same coverage said the Seed team is pre-training a model with a parameter scale of up to 10 trillion, on a level comparable to Anthropic's most advanced models. Later the same day, 36Kr relayed The Paper's report and confirmed Zhang's remarks on both the timeline for catching up and the state of Seed's pre-training data.

Zhang had previously said in internal meetings that model development should follow a long-term, delayed-gratification approach rather than trading others' outputs for a short-lived place on leaderboards. Public information on the matter currently remains at the stage of these internal statements, with no further developments disclosed.