ModelBest, a startup born out of Tsinghua University’s prestigious Natural Language Processing lab, has officially entered the global unicorn club. Following a Series B funding round that raised several hundred million yuan, the company’s total first-quarter haul for 2026 has surpassed 1 billion yuan. This influx of capital positions the firm as a leader in the specialized field of base models with full-stack self-研发 capabilities.
This latest investment round, led by Shenzhen Capital Group and Huichuan Technology, signifies a rare 'dual-city' endorsement from China’s premier tech hubs, Beijing and Shenzhen. While Beijing provided the initial research cradle through various municipal artificial intelligence funds, Shenzhen’s entry signals a decisive shift toward industrial application and hardware integration. The company's strategic move to secure both state-backed and industrial capital highlights its pivotal role in the national AI strategy.
Unlike the global arms race for massive cloud-based models, ModelBest is championing a 'Density Law' over the traditional 'Scaling Law.' By focusing on high-intelligence, low-parameter models like its MiniCPM series, the firm aims to bring sophisticated AI directly to smartphones, automobiles, and industrial robots. This 'Edge AI' approach allows for local processing, reducing latency and bypassing the need for constant cloud connectivity.
This lean technical path is particularly strategic given the current global constraints on high-end computing power. By optimizing intelligence for edge devices, ModelBest is positioning itself to lead China’s 'Intelligence Economy,' a key pillar of the 2026 national policy agenda. With government targets aiming for 70% smart terminal penetration by 2027, the startup's focus on efficiency over size addresses a massive impending market need.
Led by CEO Li Dahai, the former CTO of Zhihu, and Chief Scientist Liu Zhiyuan, a renowned Tsinghua professor, the company is also moving into hardware. Plans are underway to release AI-native development boards and specialized hardware boxes by mid-2026. This software-hardware synergy creates a closed-loop ecosystem, enabling the rapid deployment of 'embodied AI' in complex, real-world industrial and consumer scenarios.