Baichuan AI

Baichuan AI (百川智能) is a large language model company founded in 2023 by Wang Xiaochuan, the founder of Sogou, focused on Chinese-optimized general AI technology research and services. As a leading Chinese AI startup, Baichuan AI holds an important position in the LLM race with its strengths in Chinese language processing and medical domain expertise, plus its distinctive "understands search" product character.

Key Features

• LLM family: Released Baichuan-7B/13B, Baichuan 2, the over-100-billion-parameter Baichuan 3, and the latest Baichuan 4, scoring 80.64 on a comprehensive Chinese capability benchmark to set a domestic record, surpassing GPT-4-Turbo
• Baixiaoying AI assistant: Its first AI application product, integrating search with conversational interaction, supporting multi-turn search, targeted search, file interpretation, and structured information output
• Medical expertise: Outstanding performance in medical evaluations, beating GPT-4 on authoritative benchmarks such as MCMLE, MedExam, and CMExam, pursuing a "building doctors" strategic direction
• Multimodal capabilities: Supports text generation, coding, mathematical reasoning, and poetry writing, with long-text processing and role-play features
• Enterprise services: MaaS and AaaS service models with API interfaces and enterprise application integration

Use Cases In healthcare, professionals use its medical LLMs for diagnostic assistance and knowledge Q&A; in daily use, consumers experience a search-savvy intelligent assistant through the Baixiaoying app, receiving accurate information with cited sources; in enterprise settings, developers integrate Baichuan models' Chinese processing and professional capabilities through the API platform.

Technical Strengths Baichuan AI inherits the search technology strengths Wang Xiaochuan accumulated during the Sogou era, deeply fusing search capabilities with large language models to form a unique technical moat. The company applies innovative training techniques such as "importance preservation" and "dynamic data selection," and improves model performance through RLxF reinforcement learning alignment. As a Chinese-optimized LLM company, it reached a 20 billion CNY valuation within just six months of founding, demonstrating strong technical capability and market prospects.