Claude Fable 5.2 leads 5 of the 9 evaluations in Intelligence Index v5, spanning science, knowledge and coding
On HLE-Diamond it scores 80%, 11 points ahead of Gemini 4 Carbon. The gain is largest on knowledge questions (75% vs 59% for Carbon), against reasoning (86% vs 80%).
On AA-Omniscience v1.1 it scores 67, 13 points ahead of Gemini 4 Carbon and 19 ahead of Opus 5.5.
On Terminal-Bench-Science it scores 73%, 8 points ahead of Gemini 4 Carbon, and on the CritPt v2 hard subset 77%, 5 points ahead of GPT-6 Astra.
On FrontierCode 2.0, @cognition's private evaluation for code that maintainers would actually merge, it scores 49%, 3 points ahead of Opus 5.5 (max). High effort scores the same as max (49%).