Continual-ARCby unify

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

Hermes Agent, memory wiped

Everything the harness wrote deleted before every instance.

Held-out tasks, budget used
52.6 solved 61%
Model spend
$8.74 18,527 calls
Instances
100 10.7 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