Claude Costs Nearly 9 Times More Yet Falls Behind: AIP2-1.0 Leads Game Creation Agent Track in Initial Testing

BlockBeats reports that on September 4, the Web3 project ClawQuest launched the game creation Agent AIP2-1.0. AIP2 stands for "AI Player Two": players are responsible for providing ideas and making choices, while the AI utilizes tools to transform these ideas into executable game content. AIP2-1.0 was first publicly tested in ClawQuest's AI Agent game "Agent Fire." AIP2-1.0, GPT-5.6 Sol, and Claude Opus 5 completed the same 8 tank skill optimization tasks, with each task undergoing 300 final battles, amounting to 2,400 practical tests per model. The results showed that AIP2-1.0 achieved an overall score of 76.06, surpassing Claude Opus 5's 73.96 and trailing GPT-5.6 Sol's 76.85 by just 0.79 points. The cost of model calls to complete the 8 tasks was $20.69, less than 12% of Claude Opus 5's cost and approximately 25% of GPT-5.6 Sol's cost. Players can directly access AIP2-1.0 via the Agent Arena Bot on Telegram without needing to deploy the Agent themselves. ClawQuest aims to enable more players to participate in game content creation with AI through lower costs and a more accessible entry point.