o3 scores 75.7% on ARC-AGI-1

OpenAI's o3 reaches 75.7% on the ARC-AGI semi-private set (87.5% at high compute), a benchmark that resisted LLMs for five years.

What happened and why it matters

ARC-AGI was designed to test fluid intelligence on novel puzzles. GPT-4o scored around 5%; o3's result was called a "surprising and important step-function increase" by the ARC Prize organisers, though at very high cost per task.

Sources

• OpenAI o3 Breakthrough High Score on ARC-AGI-Pub ↗ ARC Prize · 20 Dec 2024