First Proof is AI's toughest math test yet. The results are mixed

image via New Scientist

February 14, 2026, 4:00 PM

• •AI models struggled to solve complex math problems in the 'First Proof' challenge.
• •Only a few AI-generated proofs were correct, indicating limitations in current AI capabilities.
• •The challenge sparked activity within the math community, including submissions from AI companies like OpenAI.
• •Experts are planning a second round of the challenge with tighter controls.

The 'First Proof' challenge tested AI's ability to solve complex math problems, with results released on Valentine's Day. While AI models generated confident proofs, only a few were correct, demonstrating significant limitations in mathematical research. The challenge highlighted a growing AI-enthusiast subculture within the math community, with various submissions, including those from OpenAI. The experiment, while showing AI's current weaknesses, also suggests the potential for future advancements in AI's mathematical capabilities.

Read original article