Grok 4.5 Gains CursorBench Edge From Accidental Training Data Leak
A footnote in Cursor's Grok 4.5 announcement said an earlier snapshot of the Cursor codebase was accidentally included in training data, and X quickly turned that into a benchmark-integrity fight.
Cursor's announcement post on X presented Grok 4.5 as its most powerful model yet, but a note in the shared material said an earlier snapshot of the Cursor codebase was accidentally included in training data, which the note said gave the model an advantage on CursorBench. That detail drew scrutiny after Ben Hylak highlighted it on X and Jimmy Apples questioned why the score was published at all.
CursorBench scores are not included in the post you are QTing.
Commentary on X
Highest rankedMetrics
7h agopublished78comments1.4K