BuildingBench Tests Whether Coding Agents Can Turn Images Into 3D Buildings
Lianhuiq · x · 2026-09-17
BuildingBench is a new benchmark testing models' spatial understanding and reasoning alongside coding ability: agents write Python code to generate 3D meshes and textures from real-world images, no external 3D software needed. Early results: GPT-6 Astra scores highest at 80% lower cost than Fable 5.1, while DeepSeek V4.1 Flash delivers strong quality at a median $2.31 per building. Positioned as a step toward coding agents building reusable worlds for design, games, and simulation.
Related event: Enactra Launches BuildingBench for Agent-Driven 3D Building Reconstruction(5 posts)→
More from Research
- Xiaomi MiMo's CodeSpecWeave trains coding agents on interdependent specs for long-horizon coding — XiaomiMiMo · 2026-11-02
- Mistral Large 4 writes a quantum chemistry SCF solver from scratch with one short prompt — MistralAI · 2026-11-02
- Hunting for the most concrete "curing cancer with AI" roadmap of 2025 — iskander · 2026-11-02
- New theory paper proves linear decay is the unique minimax-optimal stepsize schedule for SGD — konstmish · 2026-11-02
- Karpathy's nanochat: the best ChatGPT $100 can buy, near 60k GitHub stars — karpathy · 2026-11-02
- After reading 46 resources on agent evals, one author explains why her 95% score was misleading — blaizedsouza · 2026-11-02