UniPat AI
Category: AI Infrastructure
Chinese AI training-data and benchmarking startup building realistic evaluation scenarios for AI models and agents. UniPat AI was founded in 2025. The company is led by Kuan Li (李宽). Based in Beijing, China. Total funding raised: $300M. Latest round: Series A. Key investors include Alibaba Group (lead), Tencent, Sequoia China.
- Founded
- 2025
- Headquarters
- Beijing, China
- Total funding
- $300M
Value proposition
Accelerates real-world AI automation by creating realistic scenarios, training data, and benchmarks that help models become more capable, productive, and trustworthy.
Products and solutions
BabyVision multimodal visual-reasoning benchmark, UniScientist research-agent training paradigm/model, Terminal-X coding-agent benchmark, Vibe-Coding Arena for multi-turn coding-agent evaluation.
Unique value
Focuses on realistic, economically meaningful training and evaluation environments rather than building a general-purpose foundation model itself.
Target customer
Frontier-model research teams and enterprises developing AI models and agents.
Industries served
AI foundation models, agentic coding, multimodal AI, scientific research automation.
Technology advantage
Research-led human-LLM data-production approach: LLMs generate cross-disciplinary content at scale while human experts verify it; applies to agentic scientific-research and real-world agent benchmarks.
How they differentiate
UniPat positions around China-focused post-training data and scenario-based benchmarking for agents and multimodal models, whereas Scale AI and Surge AI are primarily known for data-labeling and human-feedback services and Mercor for expert talent/data workflows.
Main competitors
Scale AI, Surge AI, Mercor.
Key partnerships
Sequoia China xbench co-released the BabyVision benchmark, research collaborators include Peking University contributors.
Major milestones
Founded in December 2025, co-released BabyVision in January 2026, released UniScientist in March 2026, released Terminal-X in May 2026, released Vibe-Coding Arena in July 2026, reported $300M Alibaba-led Series A at a $2.5B valuation in September 2026.
Market positioning
AI-data and model-evaluation infrastructure provider for frontier AI development in China.
Geographic focus
China, with relevance to global frontier-model developers.
About Kuan Li (李宽)
Alibaba Tongyi AI Lab DeepResearch intern, focused on post-training analysis, synthetic data and reinforcement learning; HKUST Computer Science PhD candidate; Beijing University of Aeronautics and Astronautics BEng; Institute of Computing Technology, Chinese Academy of Sciences master's degree.
Latest news about UniPat AI
More AI Infrastructure companies
Official website: https://unipat.ai