HardGeoBench Country Accuracy
Multimodal · 2025-05-08
HardGeoBench country-level accuracy for private-photo geolocation predictions. Scores are pass@1 percentages from API default settings with a simple prompt, no sampling, and no web search or tools.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 68.8 |
| Muse Spark 1.1 | 67.5 |
| Gemini 3.7 Flash | 67.5 |
| Gemini 3.8 Flash | 67.5 |
| Gemini 3.5 Flash | 66.3 |
| Gemini 3.6 Flash | 66.3 |
| Gemini 4 Argon | 63.8 |
| Claude Sonnet 5.5 | 62.5 |
| Muse Spark Open | 62.5 |
| Claude Fable 5 | 61.3 |
| Muse Spark 1.2 | 61.3 |
| Claude Fable 5.1 | 61.3 |
| Gemini 3 Flash Preview | 60.0 |
| Gemini 3.1 Pro Preview | 60.0 |
| Muse Spark 1.3 | 60.0 |
| GPT-6 Astra | 60.0 |
| Gemini 3 Pro Preview | 58.8 |
| Gemini 2.5 Pro Experimental 03-25 | 56.3 |
| Gemini 2.5 Pro | 56.3 |
| Sonnet 4.6 | 56.3 |
| Gemini 2.0 Flash-Lite | 55.0 |
| GPT-5 | 55.0 |
| Opus 4.6 | 55.0 |
| o3-pro | 53.8 |
| Gemini 2.5 Flash Preview (09-2025) | 53.8 |
| Gemini 3.1 Flash-Lite Preview | 53.8 |
| Opus 4.7 | 53.8 |
| Opus 4.8 | 53.8 |
| Claude Opus 5 | 53.8 |
| Gemini 2.5 Flash | 52.5 |
| ChatGPT-4o Latest (source-unspecified snapshot) | 52.5 |
| Gemini 3.5 Flash-Lite | 52.5 |
| GPT-6.1 Sol | 52.5 |
| Muse Wonder | 52.5 |
| GPT-4.1 | 51.3 |
| o3 | 51.3 |
| Gemini 2.0 Flash | 51.3 |
| Claude Haiku 5.5 | 51.3 |
| GPT-5.1 | 50.0 |
| Opus 4.5 | 50.0 |