PlanBench-V - Element Recognition: leaderboard

Metric: Mean score (out of 2) on element recognition (map layout, textual and planning elements), perception level; answers to expert-curated questions on Chinese territorial spatial planning maps are scored 0 to 2 by a GPT-4o-mini judge (temperature 0) against structured reference answers with annotated critical points; objective items use exact match or semantic similarity; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 17 models tracked.

Top models

#ModelScore
1Qwen 3.6 Plus1.75
2Kimi K2.61.52
3Gemini 2.5 Pro1.41
4Qwen 3.6 Flash1.32
5GPT-5.41.23
6Claude Opus 4.71.19
7Qwen 2.5 VL 7B Instruct1.1
8GPT-4o1.05
9InternVL3-8B0.99
10InternVL3-14B0.93
11Qwen 2 VL 7B0.9
12Qwen 2 VL 2B0.74
13GPT-4o Mini0.66

Interactive version: theaggregate.ai/benchmark?slug=planbench-v-element-recognition · How It Works · Data refreshed daily, snapshot 2026-09-29.