Thinking-2507, as we'll call it for short, now leads or closely trails top-performing models across several major benchmarks.

A person who tested GPT-5 told The Information it outperformed Claude Sonnet 4 in side-by-side comparisons. That’s just one ...