As Silicon Valley giants broadly claim that AGI will arrive within two to three years, ByteDance co-founder Zhang Yiming offers a more pragmatic assessment: while ByteDance is capable of competing with Anthropic, OpenAI, and others, genuinely closing the gap with frontier large-model technology will take until 2029 or 2030 at the earliest. This assessment reflects where China's large-model industry actually stands. ByteDance currently faces a key shortcoming—its pretraining datasets are far smaller than those of leading international competitors—and it has drawn a clear line in the sand: no reliance on distillation shortcuts for self-developed models. That means forgoing short-term gains in benchmark scores and committing instead to deep, long-term investment in foundational self-research. Its flagship Seed model has kept a low profile, with its most recent major version update still at version 2.1 as of June 2026. By contrast, ByteDance's consumer-facing product Doubao, leveraging strengths in product operations and recommendation algorithms, has captured the largest user base in China's consumer chatbot market—but that advantage does not stem from a generational lead in underlying models.