Just now
Despite impressive advancements of large language models (LLMs) for language-related tasks, mathematical reasoning remains a weak point even for the most powerful AI models today. While LLMs excel at recognizing and replicating patterns in data, they often struggle with tasks that require genuine conceptual understanding — relying instead on heuristic shortcuts that can produce the right answer without true reasoning.
A solution that bridges this gap between current AI capabilities and true mathematical discovery could unlock breakthroughs that would impact quantitative finance, robotics, and other fields. That’s why we are excited to announce Toyota Ventures’ investment in Axiom, a company developing a foundation model capable of solving complex mathematical problems and providing the critical proofs behind its solutions.
Based in Palo Alto, California, the company was founded in 2025. Founder and CEO Carina Hong was previously a doctoral candidate in mathematics at Stanford, with a distinguished academic background that includes a master’s of science degree in neuroscience from the University of Oxford, where she studied as a Rhodes scholar, and conducted deep learning research projects at UCL Gatsby Computational Unit.
CTO Shubho Sengupta has been working on deep learning for the past 10 years; first at Baidu Silicon Valley AI Lab and then at Meta’s Fundamental AI Research (FAIR) Lab. He has worked on speech-to-text and text-to-speech models, large reinforcement learning models for playing Go, scaling deep learning model training and more recently on using LLMs to build automated software systems. Before that, he spent 10 years developing the fundamental parallel algorithms used for general purpose programming on GPUs. Together, Axiom has assembled a team of highly technical researchers to fulfill the company’s mission of building a self-improving superintelligent reasoner.
Axiom aims to convert mathematical knowledge from textbooks and journals into a machine-readable format, enabling its AI model to independently generate new problems and formally prove its own solutions. This model will act as a “System 2” reasoning engine, which refers to the development of AI models that can engage in slow, deliberate, and logical reasoning to solve complex problems. This could also be used to generate new mathematical knowledge and formulate mathematical statements that have yet to be proven.
“Mathematical reasoning has been one of AI’s toughest challenges. We’re thrilled to support Axiom’s team as they attempt to crack mathematical reasoning, which could lead to solving all reasoning.”
–Chris Abshire, principal, Frontier Fund, Toyota Ventures
Through leveraging System 2 reasoning, Axiom enables researchers to explore, test, and refine trading ideas with extreme precision. Longer term, Axiom aims to have applications for software, hardware, quantitative finance, engineering, and more.
Toyota Ventures is excited to participate in Axiom’s $64 million Seed round, led by B Capital. Frontier Fund principal Chris Abshire drove Toyota Ventures’ investment in the round. Visit Axiom’s website or Toyota Ventures’ portfolio page to learn more.