GPSBench (Applied) - Spatial Patterns: leaderboard
Metric: Score (%) on the spatial patterns (pick the geographic outlier among five named cities) task of GPSBench's Applied Track (geographic reasoning that combines coordinates or place names with world knowledge); 1,020 test samples; zero-shot from the model's own knowledge, no tools, no chain-of-thought or few-shot prompting, temperature 0; choice tasks score accuracy and numerical tasks one minus the mean absolute percentage error; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 95.4 | #145 |
| 2 | GPT-5 Mini (Low) | 95.2 | #176 (GPT-5 Mini) |
| 3 | GPT-4.1 | 94.2 | #240 |
| 4 | GPT-5.1 (Low) | 93.7 | #131 (GPT-5.1) |
| 5 | Qwen 3 235B A22B 2507 Instruct | 92.6 | #291 |
| 6 | GPT-4.1 Mini | 92.5 | #346 |
| 7 | Gemini 2.5 Flash (Non-reasoning) | 92.1 | #237 (Gemini 2.5 Flash) |
| 8 | Mistral Large 3 | 90.1 | #388 |
| 9 | Claude Haiku 4.5 | 87.4 | #271 |
| 10 | Mistral Small 3 | 85.9 | #616 |
| 11 | GPT-5 Nano (Low) | 59.4 | #415 (GPT-5 Nano) |
| 12 | Qwen 3 30B A3B 2507 Instruct | 8.1 | #464 |
| 13 | Qwen 3 14B (Non-reasoning) | 5.3 | #524 (Qwen 3 14B) |
| 14 | Qwen 3 8B (Non-reasoning) | 3.8 | #667 (Qwen 3 8B) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=gpsbench-applied-spatial-patterns · How It Works · Data refreshed daily, snapshot 2026-10-11.