GeoBuildBench - Level 4 (Hard): leaderboard
The 35 GeoBuildBench problems annotated with construction difficulty level 4 (hard) of 4. GeoBuildBench (Kim, Yang and Zhang, Peking University, 2026) asks an agent to turn Chinese plane-geometry problems into a program in a small geometry construction language, which is executed and rendered, with up to 5 rounds of revision from execution and visual feedback. Success rate in percent: a problem counts when every required object exists and every verification condition holds; higher is better.
Metric: Construction success rate (%; the agent's DSL program must create every required object and satisfy every verification condition within 5 interaction steps; 35 level-4 problems, rate = printed success count / 35). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 65.71 | #225 |
| 2 | Gemini 3 Flash | 62.86 | #159 |
| 3 | Qwen 3 VL 235B A22B Instruct | 20 | #521 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=geobuildbench-level-4-hard · How It Works · Data refreshed daily, snapshot 2026-11-12.