ARC Prize: Models scoring high on ARC-AGI-2 also score higher on ARC-AGI-3
AI summary
ARC Prize notes that a similar relationship exists on ARC-AGI-3: models that score higher on ARC-AGI-2 tend to also score higher on ARC-AGI-3. One notable exception is Gemini 4 Argon at the low reasoning setting, which scored only 42.1% on ARC-AGI-2 but reached 97.5% on ARC-AGI-3 under the provider adapter harness.
Body
On ARC-AGI-3, there is a similar relationship: models that score higher on ARC-AGI-2 tend to also score higher on ARC-AGI-3.
One notable outlier is Gemini 4 Argon at low reasoning, which scored only 42.1% on ARC-AGI-2 but 97.5% on ARC-AGI-3 with the provider adapter harness. https://t.co/JQzMGRjJoq