I Gave the Hardest Cryptic Crossword I Could Find to a Bunch of LLMs

The article describes an informal test of several large language models on difficult cryptic crossword clues from an Australian puzzle known for its complexity. The author compares the performance of ChatGPT, Claude Sonnet 5, and Oreate, concluding that the models showed some ability but struggled with harder clues and wordplay. The piece frames the results humorously, with the author claiming human solvers still have an advantage in this niche puzzle format.

Summary generated July 28, 2026. AI summaries can make mistakes.