Kimi K3 在制作自玩复古游戏任务中，以 18.4K tokens、$0.28 的成本击败 Opus 4.8（21.3K tokens、$0.54），成本减半。Kimi K3 的 Q*bert 和 Battle City 逻辑完整可玩，而 GPT-5.6 虽纹理最佳但游戏逻辑失效。
Kimi K3 just beat Opus 4.8 at building retro games for half the cost.
Atomic Chat, a desktop app that runs LLMs locally, asked each model to build Road Fighter, Battle City, and Q*bert as self-playing HTML files.
To make each result work, the graphics, enemies, game rules and a player that could play on their own all had to be in the same file.
Because a single error can ruin the entire simulation, setup tests are more important than code generation.
Outputs:
Kimi K3: 18.4K tokens, $0.28
GPT-5.6: 18.1K tokens, $0.28
Opus 4.8: 21.3K tokens, $0.54
GPT-5.6 generated the best looking textures, but the logic for the cars/tanks did not work. Shows how great looks can be deceiving for poor control and state management.
Kimi's best work was Q*bert, where the agent changed the state of the cube, moved around the pyramid, and avoided an enemy, without breaking the game loop.
New Kimi K3 beat Opus 4.8 at making retro games for 2x cheaper! We gave three models the same task: make three old arcade games that play by themselves. Each game is one HTML file with a bot that plays it Prompts: – Road Fighter – Battle City – Q*bert Outputs: Kimi K3: 18.4K tokens, $0.28 GPT-5.6: 18.1K tokens, $0.28 Opus 4.8: 21.3K tokens, $0.54 Kimi K3 made the best games. Its Q*bert is the best of all three. The player jumps on the cubes, paints them and runs from the purple snake. Its Battle City looks just like the real game and plays right. The tanks move, aim, shoot and break the walls. GPT-5.6 could not handle the cars. Its Battle City broke too. The tank died on its own and the base got hit. But GPT had the nicest textures of the three在 X 查看被引用的帖子
来源：Rohan Paul · x.com