Axiom develops an AI mathematician as the initial component of a larger self-improving superintelligent reasoner. This system focuses on advancing automated reasoning capabilities through iterative learning. The core offering is foundational AI designed for complex problem-solving and mathematical inference.
Funding
$64M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
Current AI systems struggle to perform autonomous, high‑level mathematical reasoning and to generate verifiable proofs, limiting their usefulness for scientific discovery and complex problem solving.
Solution
Axiom is developing a self‑improving reasoning engine that begins with an AI mathematician capable of tackling open mathematical problems. The platform trains large‑scale transformer models on curated math datasets, then applies hierarchical planning, formal verification, and self‑play loops to iteratively refine proof‑generation skills. Successful conjectures and proofs are fed back into the training pipeline, creating a continuous improvement cycle that accelerates discovery. Axiom provides programmatic APIs and research toolkits so external teams can integrate the engine into domain‑specific reasoning workflows. The architecture is built for distributed compute clusters, enabling scalable training and rapid model updates without manual intervention. This foundation is intended to serve as a stepping stone toward broader superintelligent reasoning capabilities.
Target Audience
Primary users are academic research labs, university mathematics departments, AI research organizations, and enterprises that require advanced automated reasoning for scientific or engineering challenges.
Features
- Transformer‑based models pre‑trained on extensive mathematical corpora, including unsolved conjectures such as the Collatz problem
- Hierarchical planning module that decomposes complex proofs into tractable sub‑tasks
- Formal verification layer that interfaces with proof assistants (e.g., Lean, Coq) to ensure logical soundness
- Self‑play reinforcement loop where the system generates, tests, and critiques its own conjectures to drive continual learning
- Automated generation of novel mathematical objects linked to formal proof pipelines
- Scalable distributed training infrastructure supporting multi‑node GPU clusters and dynamic resource allocation
- Open‑source research SDK with Python bindings, enabling custom dataset ingestion and experiment tracking