On May 26, Sholto Douglas, an engineer at Anthropic, announced on the X platform that their latest model, Claude Mythos, successfully solved a century-old problem in combinatorial geometry—the Erdős unit distance conjecture—through a "clever and concise proof." This breakthrough followed OpenAI's recent achievement of overcoming this milestone mathematical problem using GPT-5.5, marking the intensifying competition among leading AI vendors in the field of pure mathematical discovery.

QQ20260527-091235.jpg

This breakthrough demonstrated a collaborative agent paradigm different from traditional single models. According to reports, the team built an automated testing system containing multiple independent Claude Code instances. Within this system, agent instances with access to Mythos were responsible for receiving problems and generating different solution paths, which were then summarized and distributed to other independently running instances for cross-validation.

Although mathematician Daniel Litt commented that Mythos' initial results were "slightly inferior" to OpenAI's solution, the model ultimately found the same problem-solving path as OpenAI and independently derived a more original and concise proof. Currently, Anthropic has officially released a proof version compiled by Opus4.7.

This development has once again sparked industry discussions about the development paths of "pure large language models (LLMs)" and "symbolic reasoning assistance systems." Previously, Google DeepMind had announced that it had solved nine Erdős problems using its AlphaProof Nexus system, but due to its high dependence on the Lean formal proof language, it triggered debates about the value of autonomous reasoning in the purely generative AI field.

In contrast, Claude Code, as an agent framework, is now tackling humanity's top scientific challenges through multi-agent collaboration. The AI field is currently undergoing dramatic paradigm shifts, and a Fields Medal winner recently confirmed that ChatGPT5.5Pro can autonomously complete doctoral-level mathematical research within two hours. As Claude Mythos and GPT-5.5 demonstrate autonomous evolution capabilities in cutting-edge areas such as algorithm discovery and system vulnerability exploitation, AI is moving comprehensively from the early "prompt engineering" stage toward an era of autonomous agents with deep logical reasoning and scientific exploration capabilities.