ℏεsam

ARC AGI 3 benchmark has been released with a $700K prize. The benchmark focuses on exploring novel environments, acquiring goals on the fly, building adaptable world models, and continuous learning. Notably, Claude Opus 4.6, which scored 69.2% on ARC AGI 2, dropped dramatically to just 0.2% on ARC AGI 3, highlighting the significant difficulty jump between versions.

ARC AGI 3 was just released and the prize is $700K. and Opus 4.6 which scored 69.2% on high in ARC 2, just scored 0.2% on max for ARC 3. the goal of this benchmark is "to explore novel environments, acquire goals on the fly, build adaptable world models, and learn continuously."

<p>ARC AGI 3 was just released and the prize is $700K. and Opus 4.6 which scored 69.2% on high in ARC 2, just scored 0.2% on max for ARC 3. the goal of this benchmark is &quot;to explore novel environments, acquire goals on the fly, build adaptable world models, and learn continuously.&quot;</p>