Analysis
Alibaba is leading a $300 million investment in UniPat AI, an AI training and benchmarking startup, at a $2.5 billion valuation, according to Bloomberg, which cited people familiar with the deal. Tencent is also participating, alongside existing backer HSG, the firm formerly known as Sequoia China.
UniPat's founder, Li Kuan, interned at Alibaba's Tongyi Lab before starting the company, and Alibaba's own framing of the deal -- a vote of confidence in a former intern -- doubles as a talent-pipeline signal to researchers still inside its labs. The company builds models, agents and benchmarks focused on scientific research and multimodal evaluation, including UniScientist, a research-grade reasoning system; SWE-Vision, a tool for evaluating code-based visual understanding; and BabyVision, which tests multimodal models against human-level visual reasoning.
The round puts UniPat in a competitive set that spans both US and Chinese evaluation vendors:
“- LMArena (formerly Chatbot Arena) -- the crowdsourced leaderboard that Scale AI's own Seal Showdown and similar tools now compete against directly.”
- Scale AI -- the incumbent US data-labeling and evaluation company, whose relationship with Meta reshaped the category over the past two years.
- Surge AI -- profitable since launch, raised its first outside capital in mid-2025 at a reported $15-25 billion valuation, pitching precision over volume to frontier labs that want an independent vendor.
- LMArena (formerly Chatbot Arena) -- the crowdsourced leaderboard that Scale AI's own Seal Showdown and similar tools now compete against directly.
Against that field, UniPat's $2.5 billion mark on a $300 million round is a fraction of Surge AI's reported range -- priced more like a fast-rising regional challenger than an established category leader, but still a large enough check that Alibaba and Tencent are effectively co-funding the evaluation infrastructure both companies will eventually be judged against.
What the announcement doesn't resolve is UniPat's revenue, customer concentration, or whether the round includes any secondary component for early employees. It's also worth asking how independent an evaluator can really be once its largest investor is also one of the AI labs whose models it benchmarks -- Alibaba's own Qwen model family is a direct beneficiary of any advantage UniPat's tools confer, the same structural tension Scale AI faced once Meta took a stake in it.
The next disclosure to watch is whether UniPat publishes benchmark methodology and funding-source transparency alongside its evaluation results, the same standard Western frontier labs have come under pressure to meet from their own investors.
The round also lands inside a broader capital shift: Chinese AI labs including Moonshot and DeepSeek have raised large rounds this year partly on the strength of homegrown benchmark performance, and a well-funded, more neutral scoring layer gives both investors and regulators a reference point that isn't produced by the labs marketing their own results. Alibaba and Tencent each have obvious reasons to want that reference point to exist -- and obvious reasons to want influence over how it's built.
