ByteDance CEO Liang Rubo admitted at the company’s 2026 mid-year all-hands meeting on August 6 that the gap between its large language models (LLMs) and those of leading overseas competitors is widening. While ByteDance remains competitive in consumer-facing applications and video generation, it faces severe pressure in the core LLM race, a foundational track for Artificial General Intelligence (AGI). In the first half of 2026, Doubao maintained strong competitiveness in consumer applications, and the video generation model Seedance holds a State Of The Art (SOTA) position. However, in the LLM domain, the technological gap relative to top overseas products is showing signs of widening. Liang said ByteDance will refuse to take shortcuts and will not package products using external models to chase short-term benchmark rankings. He stressed the need for "delayed gratification" and sustained in-house R&D investment. "On large language models, ByteDance will continue to insist on self-research going forward, do the basics well, accept being behind in the short term, and persist in long-term optimization," Liang said. He indicated this persistence is based on the company’s forecast of future technology trends. Recently, Doubao, Feishu, and Volcano Engine completed a strategic alignment to break down departmental silos and promote AI in office collaboration and enterprise services. Liang revealed that over 90% of new Feishu clients simultaneously purchase Feishu’s AI-related products