Today, Logical Intelligence announced that its Aleph tool achieved a 76 percent score on the Putnam Benchmark, one of the most demanding mathematical reasoning tests in artificial intelligence. The benchmark measures a model's ability to solve formal mathematics problems by producing verified proofs rather than relying on text generation. While Aleph is an internal tool built on top of an LLM, its performance places it ahead of all publicly evaluated LLMs and the hybrid EBM systems that still depend on LLM scaffolding. The results are a strong signal that native EBM architectures offer a clear path to trustworthy AI.

"We built Aleph as an internal tool to test the mathematical rigor of the environment we are creating, not to be our core model," said Eve Bodnia, founder and CEO of Logical Intelligence. "Aleph's performance proves that our foundations are strong, even though Aleph itself was developed on top of an LLM. The tool represents a fraction of what we expect our core model to accomplish."

Why Logical Intelligence Uses EBMs Instead of LLMs

Most AI systems reason the same way they write: one word at a time. This produces long, fragile chains of tokens that can fall apart with a single incorrect step. The model receives a "final grade" only at the end of the chain, with no idea where the reasoning failed. This makes LLMs unpredictable and unsuitable for environments that require guaranteed correctness.

Logical Intelligence uses EBMs because they operate on a different principle. An EBM does not think in words. It reasons in continuous mathematical states shaped by the structure of the problem. Instead of producing text token by token, the model updates its entire internal state at once. This allows it to correct course, explore alternatives, and converge on stable, verifiable answers. The system behaves closer to a trained mathematician than a predictive text engine.

EBMs are positioned to become the backbone of the systems where uncertainty is unacceptable. These include true self-driving vehicles, advanced aviation, automated manufacturing, power grids, defense systems, autonomous robotics, chip design, and national infrastructure. Any environment that depends on logic behaving the same way every time will require the type of deterministic reasoning that EBMs can provide.

"If you need certainty, you cannot rely on word prediction," Bodnia said. "You need a system that works through the structure of a problem. EBMs give us the foundation for that."

Why Aleph Matters

Aleph was created for one purpose. It is a tool that converts mathematical problems into formal statements and generates proofs that can be checked by a machine. This allows researchers to verify that an answer is mathematically correct. Even as an internal tool built on an LLM, Aleph's ability to generate large volumes of verifiable proofs is a meaningful advancement. Most AI systems can describe mathematics. Very few can prove anything.

"Aleph gives us a new level of certainty in AI today," Bodnia said. "It is the first signal of what is possible when you build systems around mathematical truth."

Logical Intelligence is already working with a small group of organizations to test early applications of Aleph in controlled environments across key vertical industries. These pilots are designed to explore how mathematical verification can support real systems.

Logical Intelligence will release its general purpose model with formal machine verifiable reasoning in 2026. This system will go far beyond Aleph and demonstrate how mathematical reasoning can support complex, high-assurance environments at scale. The company will show how its approach can serve industries where perfect logic is the requirement.

"Aleph is our first milestone," Bodnia said. "The full system is coming in 2026."

For more information and to read the Aleph white paper, visit www.logicalintelligence.com/aleph-prover.html.

About Logical Intelligence

Logical Intelligence is an artificial intelligence research company building the first fully language-free, mathematically grounded Energy Based Models. These systems differ from LLMs and hybrid EBM approaches by reasoning directly in structured state space and generating proofs that can be checked for correctness. Logical Intelligence is designing its models to underpin critical infrastructure, advanced automation, and high-reliability computing. Its team includes researchers with advanced degrees in mathematics and computer science, ICPC and IMC medalists, contributors to major proof systems, a Fields Medalist, and a Turing Award laureate who guides the company's long-term scientific direction. For more information, visit www.logicalintelligence.com or follow us on X at @logic_int and our founder & CEO at @EveLovesOlive.

View source version on businesswire.com: https://www.businesswire.com/news/home/20251202089385/en/

Contact

[email protected]

© 2025 Business Wire, Inc.Disclaimer:This press release is not a document produced by AFP. AFP shall not bear responsibility for its content. In case you have any questions about this press release, please refer to the contact person/entity mentioned in the text of the press release.