Comparative Analysis of Domestic Open-Source Large Language Models
Haifeng Li, Hongwei Si, Mo Hai · Procedia Computer Science · 2025
This study analyzes the development of Chinese open-source large language models (LLMs) since ChatGPT’s release in 2022. Chinese LLMs have evolved through four stages: initial exploration (1990s-2017), technology learning (2017-2023), rapid development (2023-2024), and current maturation. The research examines major models including DeepSeek, Qwen, Baichuan, ChatGLM, and Yi, with particular focus on DeepSeek-R1’s superior performance in reasoning tasks. Benchmark tests show Chinese models increasingly closing the gap with international counterparts. Two key trends emerge: growing emphasis on specialized reasoning capabilities and development of industry-specific models for sectors like finance and healthcare, indicating a shift from general-purpose to domain-specialized applications with practical value.