Grok 4.7 jumps to #3 on BuildingBench 3D generation, 66% cheaper than Fable 5.1
ZhitingHu · x · 2026-09-22
Third-party evals show Grok 4.7 (xhigh) scoring 0.783 on BuildingBench, up from 0.696 for 4.6 (+12.5%), leaping from #9 to #3—behind only GPT-6 Astra (0.843) and Fable 5.1 (0.814). Missing walls, inside-out surfaces and blurry textures are largely fixed. Median cost per building: $11.43 vs Astra's $7.11 and Fable's $33.15.
Related event: Grok 4.7 jumps to third on BuildingBench 3D generation benchmark(2 posts)→
More from Multimodal
- New Qwen-Image-2.1 LoRA blends styles from two input images — bdsqlsz · 2026-09-22
- Grok 4.7 vs 4.6 Blender render comparison shows clear gains in shadows and texture — mattyp · 2026-09-22
- Pika unveils new AI creative platform with Relight tool that re-lights any photo or video — aziz4ai · 2026-09-22
- AI Short Film 'Sleep Paralysis Demon' Made with MiniMax H3 + SeedVR2 — Warp_d · 2026-09-22
- Face Swap with Qwen Image 2.1: No Masking, Just Two Images and a Prompt — sci032 · 2026-09-22
- MiniMax H3 Max Tops Design Arena Video Editing Leaderboard with Elo 1373 — noahsolomon · 2026-09-22