From 0% to 36% on Day 1 of ARC-AGI-3
4 hours ago
- #AI
- #Symbolica
- #ARC-AGI-3
- Agentica SDK by Symbolica scores 36.08% on ARC-AGI-3, completing 113 out of 182 playable levels.
- It outperforms CoT baselines (Opus 4.6 Max: 0.2%, GPT 5.4 High: 0.3%) at a lower cost ($1,005 vs. $8,900).
- The SDK completes 7 out of 25 available games and is available on GitHub.
- Agentica SDK is sandboxed for persistent tasks, including solving ARC puzzles.
- Discrepancy noted in human baseline scores for game cn04 levels.