Parakhin on GPT-5.6 Math: Better Than Fable, Worse Than 5.2 Pro

MParakhin · x · 2026-07-20

Microsoft CEO Mikhail Parakhin responded to developer feedback regarding model math capabilities, stating that GPT-5.6 indeed performs significantly better than Fable (especially the Pro version) in math research, but still falls short of the legacy GPT-5.2 Pro overall. He speculates that GPT-5.2 Pro's astonishing reasoning ability stems from its massive reasoning budget and the likely use of a large-scale best-of-n sampling strategy.

Related event: GPT-5.6 Solves Decades-Old Math Problems, Boosting Proof Capabilities(10 posts)→

Original post →

More from Models

Models channel →