SWE-Game Benchmarks 247 Godot Build Tasks, Claude Opus 5 Tops Scores but Stays Below 60/100

SWE-Game evaluates coding agents on 247 tasks built from 41 executable Godot reference games across 13 gameplay categories: development from briefs, implementation from design docs, skeleton completion, repair of 83 injected faults, and Godot-to-Unity porting. Claude Opus 5 was the top performer but best construction scores stay below 60/100, with the brief-to-game task topping out at 50.38. Rubric-based visual scoring hit 0.829 Spearman correlation with human raters.