PlanBench-V - Scheme Evaluation: leaderboard

PlanBench-V (2026) tests vision-language models on interpreting Chinese and international spatial planning maps with questions curated by professional planners; answers receive scores from 0 to 2 against reference answers with annotated critical points, using objective checks or a GPT-4o-mini judge. This board is the mean score on judging the strengths and weaknesses of a planning proposal; higher is better.

Metric: Mean score (out of 2) on judging the strengths and weaknesses of a planning proposal, implementation level; answers to expert-curated questions on Chinese and international 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 February 2027. 17 models tracked. Updated 30 Sep 2026.

Top models

#ModelScoreOverall rank
1Qwen 3.6 Plus1.62#226
2GPT-5.41.59#128
3Kimi K2.61.44#181
4Gemini 2.5 Pro1.44#269
5Claude Opus 4.71.43#78
6GPT-4o1.22#744
7Qwen2.5-VL-72B-Instruct-AWQ1.15#706
8Qwen 3.6 Flash1.05#431
9InternVL3-9B0.921#1703
10Qwen 2 VL 7B0.857#2324
11Qwen 2.5 VL 7B Instruct0.802#1867
12InternVL3-14B0.709#1204
13Qwen 2.5 VL 3B Instruct0.697#2562
14Qwen2-VL-72B-Instruct-AWQ0.67
15GPT-4o Mini0.636#1658

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=planbench-v-scheme-evaluation · How It Works · Data refreshed daily, snapshot 2026-11-16.