Continual-ARCby unify

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

Program library

Every answer is a program, kept once it reproduces the demonstrations and run first when the task returns.

Held-out tasks, budget used
26.3 solved 89%
Model spend
$0.74 550 calls
Instances
100 3.8 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.

Variant
Effort
Order

Loading the run