GeoAgentBench (Plan-and-Solve): leaderboard
Metric: Parameter Execution Accuracy (%): share of ground-truth steps whose last matching tool call has semantically equivalent parameters, on 53 spatial analysis tasks over 117 atomic GIS tools in an executable sandbox (30 steps per task), agents under the Plan-and-Solve paradigm (a full plan first, then execution without feedback); higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Flash | 39.86 |
| 2 | DeepSeek V3 | 37.82 |
| 3 | GPT-4o | 36.95 |
| 4 | GPT-4o Mini | 30.74 |
| 5 | Claude Sonnet 4.6 | 25.48 |
| 6 | Llama 3.1 8B Instruct | 24.99 |
| 7 | Qwen 2.5 7B Instruct | 8.24 |
Interactive version: theaggregate.ai/benchmark?slug=geoagentbench-plan-and-solve · How It Works · Data refreshed daily, snapshot 2026-10-07.