Alibaba leads $300M round for AI-testing startup UniPat at $2.5B
What's the deal? Alibaba is set to lead a $300 million late-stage investment in UniPatDealroom has a profile for this one. Try Dealroom →, an AI training and benchmarking startup, valuing the company at $2.5 billion. Tencent and previous backer HSGDealroom has a profile for this one. Try Dealroom →, formerly Sequoia China, also joined the round, according to people familiar with the matter.
What's the endgame? UniPat designs realistic scenarios to test and measure AI models, then sells the resulting training and benchmarking data to researchers and businesses. It hosts benchmarks that evaluate AI coding agents, browser agents, and multimodal models on visual reasoning.
Who's behind it? Founder Li Kuan worked in Alibaba's Tongyi AI lab last year on post-training analysis, data synthesis, and reinforcement learning. His lab, founded in late 2025, also builds smaller models for prediction and scientific research, with backing from Monolith and ByteDance-backed Jinqiu FundDealroom has a profile for this one. Try Dealroom →.
Why now? Large language models are running out of human-generated internet data amid rising copyright and privacy constraints. Developers also face inflated leaderboard scores that fail to reflect real-world performance. UniPat aims to address both bottlenecks.
By the numbers: The $300 million figure ranks in the 97th percentile of all-time late-stage VC rounds, based on a sample of 19,305 deals. Talks remain in progress, and the deal's terms may still change, the people said.
The signal: The round shows the premium big companies will pay to secure high-quality data on AI performance. Chinese breakthroughs this summer, like Moonshot's Kimi K3, proved their edge via global benchmarking sites. US rivals take other paths: Meta bought a $14 billion stake in data-labeler Scale AIDealroom has a profile for this one. Try Dealroom →, while Mercor — a human-expert annotation marketplace — is in talks to raise at a $20 billion valuation.
Read more: Bloomberg
Image credit: learn_tek
