Codex, Opus, Gemini Try to Build Counter Strike
13 days ago
- #AI Models
- #Multiplayer
- #Game Development
- Three major AI model updates were tested: Gemini 3 Pro, Codex Max 5.1, and Claude Opus 4.5.
- The challenge was to build a basic multiplayer 3D UI version of Counter Strike.
- Claude Opus 4.5 excelled in frontend tasks, creating visually appealing maps, characters, and guns.
- Gemini 3 Pro performed best in backend tasks, handling multiplayer and persistence with fewer errors.
- Codex Max 5.1 was a balanced performer, doing reasonably well in both frontend and backend tasks.
- Each model was given 7 consecutive prompts divided into frontend (game mechanics) and backend (multiplayer functionality).
- Claude's designs were more visually appealing, while Gemini's logical changes were more robust.
- Codex had some initial bugs but fixed them quickly, though its visuals were less impressive.
- All models successfully built a multiplayer FPS with zero hand-written code, showcasing their iterative capabilities.
- The experiment highlighted areas for improvement, such as better handling of React hooks and reducing the steep learning curve for non-programmers.