GPT-6.2 Astra tops AA-3D at 1318, ahead of Claude Fable 5.2's 1271

ArtificialAnlys · x · 2026-11-18

On AA-3D, a human-graded benchmark for building interactive 3D experiences, GPT-6.2 Astra leads with Elo 1318, followed by Claude Fable 5.2 (max) at 1271 and GPT-6.2 Sol at 1259. Kimi K3.1 Pro (1112) and GLM-5.3 (1027) sit mid-pack; MiniMax M5 scores 746, with Horizon 3 Gemini (591) and Vertexion 3 Ultimate (583) at the bottom.

Original post →

More from Models

Models channel →