GeoAgentBench (ReAct) - Tool Retrieval F1: leaderboard
Metric: Tools-Any-Order F1 (%) between the tools the agent invokes and the ground-truth tool set, on 53 spatial analysis tasks over 117 atomic GIS tools in an executable sandbox (30 steps per task), agents under the ReAct paradigm (thought, action and observation loop with runtime feedback); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 83.59 |
| 2 | GPT-4o | 80.65 |
| 3 | Gemini 2.5 Flash | 78.55 |
| 4 | DeepSeek V3 | 78.16 |
| 5 | GPT-4o Mini | 65.71 |
| 6 | Llama 3.1 8B Instruct | 47.64 |
| 7 | Qwen 2.5 7B Instruct | 28.52 |
Interactive version: theaggregate.ai/benchmark?slug=geoagentbench-react-tool-retrieval-f1 · How It Works · Data refreshed daily, snapshot 2026-10-07.