Ex-Microsoft Exec Tests Carbon: Beats 5.2 in Math, Unreliable for Coding

Former Microsoft executive Mikhail Parakhin found Carbon clearly beats Fable 5.2 in math and ML reasoning at the High tier, but still can't be trusted for coding; he suggests raising reasoning effort to let it self-check longer on tricky bugs.

2026-11-12 ~ 2026-11-12 · 3 related posts

• Ex-Microsoft exec benchmarks Carbon vs 5.2: Carbon wins at math, still distrusts its coding — MParakhin · 2026-11-12
• Ex-Microsoft CTO: Carbon often beats Fable 5.2 on math, but he still won't trust it with coding — melvinjohnsonp · 2026-11-12
• MParakhin's tip for coding with Carbon: raise reasoning effort and let it review changes longer — MParakhin · 2026-11-12