返回全部资讯
Rohan Paul (@rohanpaul_ai)T2·基座大模型

Fable 5.2 与 GPT-6.1 体素建模成本差 8 倍

原标题:Interesting experiment by AI/ML API (@aimlapi).
42AI 研判分

AI/ML API 让 Claude Fable 5.2 与 GPT-6.1 Sol 分别用 3D 体素重建《赛车总动员》活塞杯决赛场景,前者花费 108.87 美元,后者仅 13.02 美元。成本差异源于 Fable 5.2 严格遵循体素风格,连烟雾都逐块建模,而 Sol 在撞车镜头中车辆仅呈暗色剪影。该对比直观展示了两模型在风格约束遵循与视觉细节上的取舍。

报道正文

Rohan Paul (@rohanpaul_ai) · 2026-11-09 21

Interesting experiment by AI/ML API (@aimlapi). Claude Fable 5.2 against GPT-6.1 Sol on building a 3D voxel movie scene.

Fable 5.2 spent 108.87andGPT−6.1Sol108.87 and GPT-6.1 Sol 13.02.

But that cost reflects that extra work: Fable 5.2 sticks to voxels. A strict voxel style means every puff of smoke is geometry the model has to place, and Fable did it in every shot.

Also, in the crash shot, Fable’s flipping car is lit and its paint scheme is readable. Sol’s car is a dark silhouette in the same shot.

you can try Fable 5.2 vs GPT-6.1 Sol on aimlapi[.]com

↳ 引用 @aimlapi (AI/ML API): Fable 5.2 vs GPT-6.1 Sol at Cars movie 🏁

Fable 5.2 re-created the Cars Piston Cup finale in 3D voxels for $108.87 while GPT-6.1 Sol's version cost $13.02

Run both models via API -> https://aimlapi.com

推文配图