GeoBench
GeoGuessr-style location from a photo.
Agent Index weight
0%
shown for context only
Models tested
42
from the benchmark’s leaderboard and reports
Results
| # | Model | Score | Normalised | Harness | Cost/task | Date |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5 Supported Anthropic | 89.0% | source | — | — | |
| 2 | Gemini 3 Flash Preview Supported Google | 88.0% | source | — | — | |
| 3 | Gemini 2.5 Pro Supported Google | 86.0% | source | — | — | |
| 4 | Gemini 3.8 Flash Supported Google | 86.0% | source | — | — | |
| 5 | Muse Spark 1.3 Estimated Meta | 86.0% | source | — | — | |
| 6 | Gemini 3.5 Flash Supported Google | 85.0% | source | — | — | |
| 7 | Claude Fable 5.1 Supported Anthropic | 84.0% | source | — | — | |
| 8 | Gemini 3.1 Pro Preview Supported Google | 84.0% | source | — | — | |
| 9 | Gemini 3 Pro Preview Supported Google | 84.0% | source | — | — | |
| 10 | Gemini 3.1 Flash Lite Google | 83.0% | source | — | — | |
| 11 | Claude Opus 4.8 Supported Anthropic | 81.0% | source | — | — | |
| 12 | GPT-5 Supported OpenAI | 81.0% | source | — | — | |
| 13 | o1 Estimated OpenAI | 80.0% | source | — | — | |
| 14 | Gemini 2.0 Pro Google | 77.5% | source | — | — | |
| 15 | Gemini-2.0-Flash Google | 77.0% | source | — | — | |
| 16 | Gemini 1.5 Flash (Sep 2024) Google | 76.0% | source | — | — | |
| 17 | Gemini 2.5 Flash Estimated Google | 76.0% | source | — | — | |
| 18 | Claude Opus 4.5 (latest) Supported Anthropic | 75.0% | source | — | — | |
| 19 | Claude Sonnet 5 Supported Anthropic | 74.0% | source | — | — | |
| 20 | o3 Supported OpenAI | 74.0% | source | — | — | |
| 21 | Claude Sonnet 4.6 Supported Anthropic | 73.0% | source | — | — | |
| 22 | GPT-5.6 Sol Supported OpenAI | 73.0% | source | — | — | |
| 23 | GPT-4.1 OpenAI | 72.0% | source | — | — | |
| 24 | GPT-4o (2024-08-06) Supported OpenAI | 71.0% | source | — | — | |
| 25 | Gemini 2.0 Flash Thinking (Jan 2025) Google | 70.0% | source | — | — | |
| 26 | Claude 3.7 Sonnet Supported Anthropic | 68.0% | source | — | — | |
| 27 | GPT-4o mini OpenAI | 64.0% | source | — | — | |
| 28 | o4-mini Estimated OpenAI | 64.0% | source | — | — | |
| 29 | Claude-Sonnet-3.5 Estimated Anthropic | 62.0% | source | — | — | |
| 30 | Qwen2.5-72B Estimated Alibaba (Qwen) | 62.0% | source | — | — | |
| 31 | Grok 4.7 Estimated xAI | 61.0% | source | — | — | |
| 32 | Claude Sonnet 4.5 (latest) Supported Anthropic | 60.0% | source | — | — | |
| 33 | Grok 4.3 Beta xAI | 56.0% | source | — | — | |
| 34 | Gemma 3 27B Google | 52.0% | source | — | — | |
| 35 | Llama 3.2 90B Meta | 52.0% | source | — | — | |
| 36 | Llama 4 Maverick Meta | 52.0% | source | — | — | |
| 37 | Claude-Opus-4 Supported Anthropic | 49.0% | source | — | — | |
| 38 | Grok 4 Supported xAI | 45.0% | source | — | — | |
| 39 | Qwen3.8 27B Alibaba (Qwen) | 38.0% | source | — | — | |
| 40 | Claude Sonnet 4 Supported Anthropic | 37.0% | source | — | — | |
| 41 | Pixtral 12B Mistral | 37.0% | source | — | — | |
| 42 | Claude 3.5 Haiku Anthropic | 34.0% | source | — | — |