Rocket League 3D 测试：GPT-6 Astra 明显强于 Kimi K3.1

am.will @LLMJunky原推发布于 2026.10.24

在 X 上打开原推

I tested Kimi K3.1 on Rocket League Bench, and yes! It's definitely the best open model that I've tested so far. But it is easily a full generation behind the frontier on 3D modeling -- at least on this one single test. Opus 5 GPT-6 Astra Fable 5.1 Opus 5.5 All provided significantly better results. This test cost around $11 total, decent, but at only half the price of Opus (and probably less token efficient), I don't think I'd ever reach for this (for this type of task). It is good, not excellent.