Milkyway Andromeda collision by Opus, Fable and GPT-6 Astra. One model won and it is gorgeous.
一位用户提示 Fable 5.1、Opus 5 和 GPT-6 Astra 模拟银河系与仙女座星系的碰撞。每个模型大约花费了 5 美元的代币。用户认为 Opus 5 的模拟效果优于 Fable 5.1，而 GPT-6 Astra 的结果最差。这些模拟结果已发布在 GitHub 上供查阅。
时间与来源
时间显示为 UTC
显示时区：UTC
本地时区尚不可用，暂时显示 UTC。
发布当时偏移：UTC+02026年9月12日 09:32 UTC
收录当时偏移：UTC+02026年9月12日 11:01 UTC
- 发布
- 2026年9月12日 09:32
- 收录
- 2026年9月12日 11:01
- 来源类型
- 开发者社区
- 档位
- 社区
- 信源状态
- 正常
档位是按信源手工设定的编辑判断，不是逐条打分。
I told Fable 5.1, Opus 5 and GPT-6 Astra to simulate the collision of Milkyway and Andromeda via this concise prompt: "Create a simulation of the milkyway andromeda collision."
They all burnt approximately similar $ in tokens, around 5$ each. What surprises me, Opus 5 performed in my opinion better than Fable 5.1. GPT-6 Astra delivered by far the weakest outcome.
While Opus and Fable where mainly focusing on the simulation of the particles by creating a single javascript file, was GPT-6 Astra going in the direction of setting up a react project with all the boilerplate.
The physics simulation of GPT-6 Astra is very static, while both Opus and Fable created a very dynamic and accurate simulation. I cannot say that Opus or Fable did a more realistic looking particle simulation of the moving galaxy stars, but to me Opus simulation just looks better.