X(
Public post
Xavier (Xavi) Amatriain
Sep 2026 • linkedin
Post excerpt
GPT-6 (aka Astra) performs better than humans at 62.7% on ARC-AGI-3, the most robust and complex intelligence test we have at this point. It performs a whooping 99.9% when using a more optimal harness than the standard one. For comparison, GPT-5.6-Sol (max) scores less than 10%. The term AGI is meaningless in my opinion (see https://lnkd.in/dHytz5B4), but this model is clearly super human in many dimensions. To be clear, this is not a first, but as model continue to get better we will see them overtake humans on m…
Find the people and context behind this post
Sign up for Super Carl to explore relationship context, warm paths, and relevant opportunities.