Axiom Math Formally Verifies Stacks: 16M Lines of Lean Code
By Ali Al-Montaser — Senior AI Technical Editor
Axiom Math, valued at $1.6 billion, has completed the first full formal verification of the Stacks mathematics project, converting 7,654 pages into approximately 16 million lines of machine-checkable Lean code. The company, which raised $200 million in a Series A led by Menlo Ventures, now provides a verified substrate for AI agents to build software and science without logical errors.
Executive Overview
Axiom Math, a company valued at $1.6 billion, announced on Wednesday, November 5, 2026, the complete formal verification of the Stacks project, converting 7,654 pages into approximately 16 million lines of machine-checkable Lean code. The announcement, made via the company's official blog and covered by TechCrunch AI, marks the first large-scale formal verification in algebraic geometry. The company, which raised $200 million in a Series A led by Menlo Ventures in March 2025, now provides a verified substrate for AI agents to build software and science without logical errors.
Technical Specification Table
📊 Official Data & Technical Specifications Card
| Technical Axis | Confirmed Official Data |
|---|---|
| 💰 Pricing & Usage Cost | No token pricing or subscription plans announced; the current product is an open Lean library. Company funding: $200M Series A led by Menlo Ventures (March 2025); valuation: $1.6B. |
| 🌐 Platforms & Immediate Availability | Official Lean library available to the research community and agents; no announcement of Web, API, AWS, Azure, or Google Cloud availability yet. |
| ⚡ Performance & Speed Metrics | 7,654 original pages → ~16 million lines of Lean. Parallel operation of an AI agent fleet with a flexible hierarchical structure. Training of internal models specialized in mathematics. |
| 🛡️ Security & Breach Resistance | Formal verification ensures logical soundness of proofs line by line, preventing contradictions and conflicting definitions. No data on prompt injection resistance because the product is not a chat interface. |
| 🧠 Context Window | Not applicable; the product is not a language model with a context interface, but a proof verification system based on the Lean proof assistant. |
| 🌍 Arabic Language & Regional Support | No official mention of Arabic language support or service availability in Arab countries; outputs are mathematically neutral Lean code. |
Deep-Dive Features & Architecture
The Stacks project began in 2005 under Johan de Jong at Columbia University as an open-source reference and expanded to include results from undergraduate algebra through algebraic geometry and number theory. The Axiom Math team faced an architectural challenge: maintaining the coherence of all theorems and definitions across 16 million lines of Lean code, as any flaw in a single definition could undermine the soundness of the entire structure. Carina Hong, the 25-year-old CEO, stated: "The big things and the delicate things in life don't agree together," referring to the sensitivity of maintaining logical consistency.
To overcome this, the company gave its agents a flexible hierarchical structure instead of full control over every step, allowing agents to discover tasks themselves and work in parallel. Axiom Math also trained its own models on mathematics, an approach Hong believes most other verification companies have not taken. Initially, the team relied on an in-house Lean expert to instill "human taste" into the agents, and then the agents gradually acquired this taste themselves.
The result fills a massive gap in formally verified mathematics. Ashvin Swaminathan, a technical staff member, stated: "When you look at what has been formally verified in mathematics, you find a huge black hole covering algebraic geometry." Now researchers have a solid foundation to check their new work in this field using Lean.
Benchmark & Competitive Performance
No numerical benchmarks comparing Axiom Math to direct competitors have been published, but the context reveals the difference: OpenAI alone announced 719 manuscripts claiming to solve difficult mathematical problems in one month, without logical guarantees. In contrast, Axiom Math delivered 16 million lines of machine-checkable Lean, equivalent to converting an entire 7,654-page library into a verified proof structure. The team cites Fields Medalist Terence Tao: "The more proofs that are formally verified, the more proofs can be produced." This positions Axiom Math as infrastructure rather than a direct competitor in language models.
Industry Impact & Enterprise Adoption
The verification of Stacks provides a foundational resource for AI agents in software engineering, finance, and scientific research. By offering a machine-checkable mathematical substrate, Axiom Math enables enterprises to build systems with provable correctness, reducing testing and auditing costs. While no commercial platforms or pricing have been announced, the open Lean library is available to researchers and developers, and the company's $200 million funding signals strong investor confidence in formal verification as a service.
Conclusion
Axiom Math's formal verification of the Stacks project represents a milestone in applying AI to rigorous mathematics. With 16 million lines of verified Lean code, the company has created a substrate that could accelerate the development of trustworthy AI systems. As the technology matures, it may become a standard for ensuring logical soundness in critical software and scientific discoveries.
Media Source: TechCrunch AI | Fact Verification & Analysis: AI Tools Oasis
Frequently Asked Questions
Axiom Math is valued at $1.6 billion and raised $200 million in a Series A round led by Menlo Ventures in March 2025, the same period the company was founded by 25-year-old Carina Hong after she left Stanford University's mathematics PhD program.
The Stacks project is one of the largest human-written mathematical collections, started by Johan de Jong at Columbia University in 2005 as an open-source reference covering algebra, geometry, and number theory. It spans approximately 7,654 pages, which Axiom Math converted into about 16 million lines of machine-checkable Lean code.
The technology rewrites mathematical proofs in a programming language called Lean, then a proof assistant program checks them line by line. If a proof passes, it becomes logically guaranteed. Axiom Math used a fleet of AI agents operating in parallel with a flexible hierarchical structure and trained its own models on mathematics to maintain definition consistency across 16 million lines.
The official announcement does not mention any support for the Arabic language or service availability in Arab countries. The product currently targets the global mathematical community and software developers, and its primary output is mathematical Lean code, a programming language not tied to any user-interface language.
The announcement did not specify commercial platforms such as Web, API, AWS, or Azure. The current product is an open official Lean library usable by agents and researchers, and the company stated that the Stacks verification is now available as a substrate on which AI agents can build.
AI Tools Oasis Team
Bringing you the latest news and analysis in the world of Artificial Intelligence with accuracy and credibility. Follow us for all updates.
