Continual-ARCby unify

ReplayARC-AGI-1 · Fresh · GPT-6 Luna, high

Prime Agent, as shipped

Every learning mechanism the harness ships switched on.

Held-out tasks, budget used
20.3 solved 98%
Model spend
$0.92 718 calls
Instances
100 3.2 hours

Budget used is the share of the cap of 16 that an instance was charged, averaged over the instances: a solved instance costs 1 for each wrong answer and each demonstration, an unsolved one all 16, so lower is better.

Loading the run