GPT-5.5’s verbosity is its secret weapon against Claude Opus

On the DesignArena leaderboard for AI game development, OpenAI's GPT-5.5 has taken the top spot with an Elo rating of 1362, just barely beating Claude Opus 4.7's 1352.

What's interesting is why GPT-5.5 won: it's because the model writes verbose frontend code — "frontend verbosity," as the benchmark calls it. That seemingly bloated coding style actually produces the most feature-complete games every time. It's a different approach from competitors like Zhipu AI's GLM-5V-Turbo, which beats Claude at generating code from screenshots. Game development has become a clear stronghold where OpenAI consistently outruns Anthropic.