Exclusive | $200M+ Valuation in Three Months! Gen Z-Founded LLM Startup Raises Another Tens of Millions in USD Funding
"Waves" has learned that Sapient Intelligence has raised another tens of millions of dollars in a Seed+ round, less than three months after its previous eight-figure dollar funding. This round's investors include JAFCO, Japan's largest private equity investment group, along with other well-known international investment institutions. Minerva Capital continues to serve as the long-term exclusive financial advisor.
By | Muxin Xu
Waves has learned that Sapient Intelligence has raised an additional tens of millions of dollars in Seed+ funding, just under three months after its previous multimillion-dollar seed round. This round's investors include JAFCO, Japan's largest private equity investment group, along with other well-known international investment institutions. Minerva Capital continues to serve as the long-term exclusive financial advisor.
We previously covered Sapient and its founder, Guan Wang, a post-00s Tsinghua graduate, in "The Gen-Z Tsinghua Kid Taking on GPT." Sapient is dedicated to developing a general-purpose large model for solving complex problems, aiming to bypass the theoretical limitations of LLMs and enable AI to reach the L2 stage faster — that is, to possess long-range reasoning capabilities and even think and self-evolve like humans.
Austin, co-founder of Sapient Intelligence, said that since the model's development began, it has hit several milestone markers, particularly in its self-learning capabilities on open-ended math problems. On expert-level Sudoku, for instance, the model achieves over 95% accuracy — something impossible for GPT-architecture-based models. Austin told us that the advantage in speed and accuracy comes down to this: "Other models essentially distill Sudoku solutions into the model through massive amounts of CoT data. Sapient doesn't need any intermediate steps or CoT data."
Currently, Sapient has completed the pre-training of its first new model architecture. The new model, Sapient H, has been tested on complex constraint satisfaction problems, high-dimensional discrete variable optimization problems, and mathematical reasoning tasks — outperforming the deeply network-optimized Llama DeepNet 100L model in validity, accuracy, parameter efficiency, and overall performance.
On the team front, Sapient has established dual R&D centers in Beijing and the Bay Area. New technical hires include senior researchers from Tsinghua and Peking University, core members from DeepSeek and reinforcement learning pioneer Richard Sutton's team, among others.
Image source | Official screenshot from Detroit: Become Human
Layout | Hongyu Liu
