How enabling two settings tripled our scores on the ARC-AGI-3 benchmark

This story was filed as a headline only — the news service holds no English full text for it. Read the original at OpenAI (RSS) →