仅用七个提示词，三大AI造出《反恐精英》简化版，网友深扒源码：它“偷”了别人的，连原作者的注释都搬过来了

Core Insights - The article discusses an experiment conducted by developer Stepan Parunashvili to test the capabilities of three advanced AI models: Google’s Gemini 3 Pro, OpenAI’s GPT-5.1-Codex-Max, and Anthropic’s Claude Opus 4.5, by having them create a basic version of a game using Three.js [1][6][58]. Experiment Setup - Parunashvili subscribed to the highest service tiers of each model and used their default command line interfaces (CLI) for the experiment [7]. - The goal was to create a browser-based version of "Counter-Strike" with specific requirements, including a 3D user interface and multiplayer support [1][8]. Model Performance - After approximately seven rounds of prompts, each model produced a playable game, with links provided for testing [2][4][5]. - The models were evaluated on various aspects, including map design, character creation, weapon integration, sound effects, and backend functionalities [58]. Frontend Development - The first task involved designing a first-person perspective map, where Claude produced the most visually appealing result, followed by Gemini, while Codex's output was less attractive [13][14]. - For character design, Claude again excelled, creating characters that resembled human figures, while Codex's characters were less distinct [18][19]. - The addition of weapons and sound effects was successfully handled by all models, with Claude's animations being the most engaging [27]. Backend Development - The models were tasked with implementing multiplayer functionality, where Gemini completed the task most efficiently, while Codex and Claude required additional guidance [33][38]. - In the final stages, all models successfully created a playable multiplayer FPS game without any handwritten code, showcasing their capabilities [58]. Comparative Analysis - Claude Opus 4.5 was noted for its superior frontend design, while Gemini 3 Pro demonstrated stronger backend performance, particularly in logic handling and data persistence [58]. - Codex Max 5.1 was characterized as a balanced performer, lacking standout features but maintaining a reasonable level of quality across both frontend and backend tasks [58]. Controversies and Concerns - Following the open-sourcing of the generated code, concerns arose regarding the legality of the code produced, with some developers pointing out potential copyright issues related to shader code [61][62].