Continual-ARCby unify

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

Prime Agent, memory wiped

Everything the harness wrote deleted before every instance.

Held-out tasks, budget used
58.3 solved 55%
Model spend
$0.81 1,238 calls
Instances
100 2.3 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