Artificial intelligence is making significant progress in the field of mathematics, solving complex problems at a surprising pace. Traditional mathematical benchmarks are struggling to keep up with these advances.

Frontier Math: a new challenge

Epoch AI has introduced Frontier Math, a rigorous benchmark designed to assess the mathematical reasoning capabilities of the latest AI tools. This test includes advanced math problems, divided into levels of increasing difficulty. The most advanced AI models, such as ChatGPT 5.2 Pro and Claude Opus 4.6, solve over 40% of the problems in the first three tiers and over 30% of the problems in the most advanced tier.

Aletheia and PhD-level mathematics

Recently, Google DeepMind announced that Aletheia, an experimental AI system derived from Gemini Deep Think, has achieved publishable PhD-level research results. Although the specific mathematical problem is niche, the result is significant for AI development. Aletheia operated in an essentially autonomous manner, without human guidance, and produced a new result.

The First Proof challenge

To address the need for more challenging benchmarks, a group of mathematicians proposed the First Proof challenge, a series of extremely difficult math problems. No one was able to provide correct solutions to all the problems by the deadline. OpenAI, with limited human supervision, managed to solve five of the ten problems.

A new frontier for AI

Epoch AI has introduced Frontier Math: Open Problems, a pilot benchmark consisting of open problems from research mathematics that professional mathematicians have unsuccessfully tried to solve. None of these problems have yet been solved by an AI. These new approaches aim to assess the capabilities of AI in mathematical areas of interest to researchers.