The company released Doubao-1.5-pro, an upgrade to its flagship AI model, which it claims outperforms OpenAI's o1 in AIME, a benchmark test that measures how well AI models understand and respond to complex instructions.