PlanBench-V - Caption: leaderboard
PlanBench-V (2026) tests vision-language models on interpreting Chinese territorial spatial planning maps with questions curated by professional planners; a GPT-4o-mini judge scores each answer from 0 to 2 against reference answers with annotated critical points. This board is the mean score on describing a planning map; higher is better.
Metric: Mean score (out of 2) on map captioning, 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: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Plus | 1.95 | #222 |
| 2 | GPT-5.4 | 1.9 | #123 |
| 3 | GPT-4o | 1.88 | #741 |
| 4 | Claude Opus 4.7 | 1.83 | #79 |
| 5 | InternVL3-8B | 1.78 | #1688 |
| 6 | Gemini 2.5 Pro | 1.78 | #266 |
| 7 | Qwen 3.6 Flash | 1.64 | #424 |
| 8 | Qwen 2.5 VL 7B Instruct | 1.63 | #1833 |
| 9 | InternVL3-14B | 1.58 | #1182 |
| 10 | Kimi K2.6 | 1.52 | #178 |
| 11 | Qwen 2 VL 7B | 1.39 | #2261 |
| 12 | Qwen 2 VL 2B | 0.93 | #6284 |
| 13 | GPT-4o Mini | 0.89 | #1623 |
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-caption · How It Works · Data refreshed daily, snapshot 2026-11-13.