SOVEREIGN view — EOSE 3-Cap Verifier at 64% leads the sovereign leaderboard.
EULER V7 (projected 52%) uses γ₁-distance scoring across 8 SOVEREIGN categories.
The gap between sovereign (64%) and frontier best (87.5%) is the semi-private set anomaly —
o3's score was on a separate 100-task eval, not the 800-task public set.
#
System
Score
Method
γ₁-dist
Year
~1000
Tasks (est.)
2025 · harder by design
4%
o3 Best
Even frontier struggles
~4%
EOSE Target
EULER V8 path · MECreature engine
~60%
Human Avg
Harder for humans too
?
SOVEREIGN Cats
Being mapped · MECreature work
ARC-AGI-2 · Leaderboard view
ARC-AGI-2 — Released 2025. Designed to resist the approaches that cracked ARC-AGI-1.
Even o3 scores ~4%. Human average drops to ~60%. The gap narrows between human and frontier —
because both are struggling. EOSE EULER path: MECreature engine + γ₁-distance scoring.
This is the unsolved game. The floor is still being poured.
#
System
Score
Method
Notes
LIVE
Interactive
Novel environments · adapt on-the-fly
30%
Jack Cole #1
Individual · Kaggle · top public
64%
EOSE Target
When submitted · MSV+CEQ
55.5%
MindsAI 2024
ARC Prize winner · $1M · PI+ENS
$50
Kaggle Budget
120 eval tasks · hard constraint
ARC-AGI-3 · Leaderboard view
ARC-AGI-3 — Interactive. AI agents must adapt on the fly to novel environments.
Not passive pattern matching — active adaptation. Jack Cole leads the public board at 30%.
MindsAI won 2024 at 55.5%. EOSE target: 64% with 3-Cap MSV+CEQ+CSE when submitted.
Cost constraint: $50 for 120 tasks. Efficiency matters as much as accuracy.