📊 BENCHMARKS & EVALS
- ARC-AGI-3 Kaggle scores jump from 7% to 56% in 30 days
Top Kaggle ARC-AGI-3 scores rose from 7% to 56% over 30 days.- Kaggle competitors using small local models and custom harnesses drove the reported gains.
- Top ARC-AGI-3 scores increased from 7% to 56% in roughly 30 days.
- ARC-AGI-3 evaluates agents on unfamiliar interactive environments under competition-specific compute constraints.
- The jump suggests competition harnesses can substantially improve performance on an evaluation designed to test human-like adaptability.