ModelBest (面壁智能) raises $700M at $2.8B valuation, becomes largest on-device AI unicorn
The AMW Read
The round confirms the on-device AI thesis as market consensus and updates the Chinese foundation-model player map, but does not invalidate open debates about commercialization.
ModelBest (面壁智能) raises $700M at $2.8B valuation, becomes largest on-device AI unicorn
Beijing-based ModelBest, the on-device foundation model startup spun out of Tsinghua University's NLP lab in 2022, has completed a financing round exceeding RMB 5 billion (~$700M), pushing its cumulative first-half 2026 funding past the same total and its valuation past RMB 20 billion (~$2.8B). The syndicate includes a national-level fund, state-owned enterprises, automakers, and prominent financial investors. The company has previously been backed by Zhihu, Huawei's Hubble, Kweichow Moutai, and China Telecom.
Why it matters: ModelBest's explosive capital trajectory — from a $5M angel round in 2022 to a $2.8B valuation in four years — signals that the on-device AI thesis has shifted from contrarian bet to market consensus. The company's self-proclaimed 'Densing Law' (published as a Nature cover article in 2025) positions its metric of intelligence density as a follow-on to OpenAI's scaling law, and its MiniCPM family of lightweight models has amassed 38 million downloads on GitHub and Hugging Face. The round also arrives as China's securities regulator expands its fifth-set listing standards to AI, opening an IPO pathway for ModelBest and its domestic peers.
Ground truth: The capital-compression arc in Chinese foundation models is reshaping the competitive map. ModelBest joins the so-called 'New Six Tigers' of Chinese AI labs, alongside DeepSeek, Zhipu AI, MiniMax, Baichuan, and others, all vying for a finite pool of sovereign and strategic capital. While the on-device segment has obvious structural advantages — lower latency, privacy compliance, offline reliability — the ability to monetize across automotive, consumer electronics, and industrial verticals at scale remains unproven. The coming 12 months will test whether density law translates into durable commercial margin, or whether the on-device thesis suffers the same commercialization drag as earlier generations of mobile AI.
#ModelBest #OnDeviceAI #ChinaAI #EndSideIntelligence #AIFunding #FoundationModels
