Baichuan 3 — это мощная языковая модель с сотнями миллиардов параметров, которая в ряде тестов превзошла GPT-4 при работе с китайским языком. Этот ИИ демонстрирует выдающиеся результаты в медицинском диагностировании, математических вычислениях и написании программного кода.
On January 29, Baichuan Intelligent officially released Baichuan 3, a large language model with hundreds of billions of parameters. In several authoritative general-purpose capability evaluations such as CMMLU, GAOKAO, and AGI-Eval, Baichuan 3 has demonstrated outstanding capabilities, especially surpassing GPT-4 in Chinese missions. And equally well in special evaluations in mathematics and code such as MATH, HumanEval and MBPP, proving the strength of Baichuan 3 in the field of natural language processing and code generation. Not only that, but its Chinese effect in authoritative medical evaluations such as MCMLE, MedExam, and CMExam, which have extremely high requirements on logical reasoning ability and professionalism, also exceeds that of GPT-4, making it the best-performing large model for Chinese medical tasks. Baichuan 3 also breaks through “iterative reinforcement learning” technology, further improves semantic understanding and generation capabilities, and performs well in the format, rhythm, and meaning of poetry creation, ahead of other large models.