This study evaluates whether heuristic learning can help AI agents rise to the top in a long-running game agent competition.
AAArena tests AI agents using fixed models to improve game code. Opus5.5 with Claude Code wins six gold medals but misses six leaderboards.