GeoBuildBench - Level 3 (Medium): leaderboard

Metric: Construction success rate (%; the agent's DSL program must create every required object and satisfy every verification condition within 5 interaction steps; 162 level-3 problems, rate = printed success count / 162). Source: arxiv.org. Saturation forecast: Around December 2026. 4 models tracked.

Top models

#ModelScore
1GPT-5.176.54
2Gemini 3 Flash72.22
3Qwen 3 VL 235B A22B Instruct34.57
4Llama 3.2 90B Vision Instruct13.58

Interactive version: theaggregate.ai/benchmark?slug=geobuildbench-level-3-medium · How It Works · Data refreshed daily, snapshot 2026-09-26.