GeoBench Photos Country Accuracy
Multimodal
GeoBench Photos Country Accuracy is the percentage of the 50 photosphere locations on the GeoBench Photos map for which the model names the correct country. Refused rounds count as misses.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 78.0 |
| Gemini 3 Flash Preview | 78.0 |
| Gemini 3.8 Flash | 76.0 |
| Claude Fable 5 | 74.0 |
| Gemini 3.5 Flash | 74.0 |
| Gemini 3 Pro Preview | 74.0 |
| Muse Spark 1.3 | 74.0 |
| Opus 4.5 | 70.0 |
| Gemini 2.5 Pro Preview 05-06 | 70.0 |
| Opus 4.8 | 68.0 |
| Gemini 2.5 Pro Experimental 03-25 | 68.0 |
| Gemini 3.1 Flash-Lite Preview | 68.0 |
| GPT-5.6 Sol | 68.0 |
| Gemini 2.5 Flash Preview 04-17 | 66.0 |
| Gemini 1.5 Flash (source unspecified) | 64.0 |
| Gemini 2.0 Flash | 60.0 |
| GPT-4o (source unspecified) | 60.0 |
| Opus 4.6 | 58.0 |
| GLM-4.6V | 58.0 |
| GPT-5 | 58.0 |
| Sonnet 5 | 56.0 |
| o3 | 56.0 |
| Qwen3-VL-235B-A22B-Instruct | 56.0 |
| Sonnet 4.6 | 52.0 |
| GPT-5.4 | 52.0 |
| Grok 4.7 | 52.0 |
| o1 (2024-12-17) | 52.0 |
| O4 Mini | 52.0 |
| Grok 4.3 | 50.0 |
| Llama 4 Maverick Instruct | 50.0 |
| Qwen2.5-VL-72B-Instruct | 50.0 |
| Claude 3.5 Sonnet (Oct 2024) | 48.0 |
| Claude 3.7 Sonnet | 48.0 |
| GPT-4o Mini | 44.0 |
| Gemma 3 27B IT | 40.0 |
| Mistral Medium 3.5 | 38.0 |
| Qwen3.8 27B | 34.0 |