Claude Opus 5 outperforms GPT-5.6 Sol on Intelligence Benchmark by almost 4x
Claude Opus 5 achieved 30.2% on ARC-AGI-3, almost 4x better than GPT-5.6 Sol’s 7.8%, and for the first time ever, we saw an AI model come up with its own algebra equations!
Claude Opus 5 achieved 30.2% on ARC-AGI-3, an AI benchmark meant to be “impenetrable”. Previously, the best score was GPT-5.6 Sol at 7.8%. But Claude Opus 5 blew it out of the water, achieving almost 4x the score. And for the very first time, we saw an AI model generate its own algebra equations!
ARC-AGI-3 was supposed to be an impenetrable barrier, but Claude Opus 5 jumped over it with ease. A difference of 4x between the top two scores means that AI models can now autonomously explore novel problem spaces in a qualitatively different way than before. This is the difference between thinking and recognizing patterns.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Articles · Latest articles