2 September 2026
AI models solved New York Times Connections puzzle nearly perfectly
First reported
The Algorithm ran this on .
- AI success rate on the daily word puzzle jumped from 18% in late 2024 to near-perfect performance by early 2025.
- The Algorithm uses Connections puzzles as a test of how AI reasoning works, revealing both what models can and cannot do.
- The newsletter invites readers to compete against AI on the same puzzles to build intuition about AI capabilities.
How it was covered
The AlgorithmMIT Technology Review
AI performance on Connections puzzles jumped from 18% success rate in late 2024 to near-perfect solving by early 2025. The newsletter uses puzzles as a window into AI strengths and weaknesses, inviting readers to test themselves against AI.