Frontier3DBench: 145 tasks probe coding agents on 3D creation, top score 66/100
Xinjie Lin · hf · 2026-11-09
Navers Lab, Einsia.AI, and Tsinghua present Frontier3DBench, a unified benchmark of 145 tasks across 9 categories evaluating coding agents on object, scene, and interactive open-world 3D creation, with joint evaluation of appearance, structure, and behavior.
Across 11 model-harness configurations, GPT-6.1-sol leads objects and scenes with 66.02/100 overall; GPT-6-astra tops interactive world construction. Key findings: every configuration scores lower on reference reconstruction than text-driven generation — plausible content is easier than reproducing specified geometry, layouts, and details. Generated worlds average 83.08/100 on local mechanisms but only 20.82 on motion clearance and 30.26 on stateful workflows, identifying reference fidelity and coordinated activity completion as central bottlenecks.
More from coding & agent
- One-person site with 27 followers made $4,800 in a day, found via Claude Code revenue scan — FinanceYF5 · 2026-11-09
- The churn era: customers now use Claude to slop up replacements for overpriced SaaS — joshgonsalves_ · 2026-11-09
- Designer builds AI pipeline that finally nails map interchange generation after 2 weeks — tomjohndesign · 2026-11-09
- AI lets designers build wild ideas, but knowing the low-level stack matters more than ever — nptacek · 2026-11-09
- Gemini CLI PR fixes orphaned child processes and wrong exit codes on headless abort — duvanka-prasad · 2026-11-09
- T3 Code hands-on: one GUI orchestrates Claude Code and Codex sessions — vista8 · 2026-11-09