StableReported 2026-10-02 12:00
Kepler: Auditable World Models for ARC-AGI-3
Kepler is an open-source tool that uses executable world models to evaluate agents in interactive environments, achieving high performance on a set of public games.
01
Evidence
- AarXiv cs.AIPrimary source2026-10-02 12:00
Kepler obtained a server-verified 100.00 RHAE on all 25 public games
- AarXiv cs.AIPrimary source2026-10-02 12:00
Abstract: ARC-AGI-3 evaluates agents in interactive environments whose rules and objectives must be inferred from observation. We present Kepler, an open-source harness that represents hypotheses as executable world models and validates them through retrospective transition checks and conditional prediction checks. Under one frozen Claude Opus 5 configuration, Kepler obtained a server-verified 10…