Multimodal benchmark
GeoBench leaderboard
As of October 2026, Gemini 3 Flash Preview has the highest published GeoBench score on Noometry at 88%, out of 25 models with results.
Last verified
About GeoBench
Guessing the country of a street-level photo, in the style of GeoGuessr.
- Category
- Multimodal
- Introduced
- 2025
- Format
- Image to location
- Unit
- Percent (random guessing ≈ 0%)
- Official site
- geobench.org
Top 15 models
- Gemini 3 Flash Preview 88%
- Gemini 2.5 Pro 86%
- Gemini 3 Pro 84%
- GPT-5 81%
- o1 80%
- Gemini 2.0 Flash (Feb 2025) 77%
- Gemini 1.5 Flash (May 2024) 76%
- Gemini 2.5 Flash 76%
- Claude Opus 4.5 75%
- o3 74%
- GPT-4.1 72%
- GPT-4o 71%
- Claude 3.7 Sonnet 68%
- GPT-4o mini 64%
- o4-mini 64%
Sponsored placements are available on pages like this one. Advertise on Noometry
All results
| # | Model | Provider | Score | Setting | Source | Date |
|---|---|---|---|---|---|---|
| 1 | Gemini 3 Flash Preview |  Google | 88% | Epoch AI | ||
| 2 | Gemini 2.5 Pro | Google | 86% | Epoch AI | ||
| 3 | Gemini 3 Pro | Google | 84% | Epoch AI | ||
| 4 | GPT-5 | OpenAI | 81% | medium | Epoch AI | |
| 5 | o1 | OpenAI | 80% | medium | Epoch AI | |
| 6 | Gemini 2.0 Flash (Feb 2025) | Google | 77% | Epoch AI | ||
| 7 | Gemini 1.5 Flash (May 2024) | Google | 76% | Epoch AI | ||
| 8 | Gemini 2.5 Flash | Google | 76% | Epoch AI | ||
| 9 | Claude Opus 4.5 | Anthropic | 75% | Epoch AI | ||
| 10 | o3 | OpenAI | 74% | medium | Epoch AI | |
| 11 | GPT-4.1 | OpenAI | 72% | Epoch AI | ||
| 12 | GPT-4o | OpenAI | 71% | Epoch AI | ||
| 13 | Claude 3.7 Sonnet | Anthropic | 68% | Epoch AI | ||
| 14 | GPT-4o mini | OpenAI | 64% | Epoch AI | ||
| 15 | o4-mini | OpenAI | 64% | high | Epoch AI | |
| 16 | Claude 3.5 Sonnet | Anthropic | 62% | Epoch AI | ||
| 17 | Qwen2.5-VL 72B Instruct |  Alibaba (Qwen) | 62% | Epoch AI | ||
| 18 | Gemma 3 27B | Google | 52% | Epoch AI | ||
| 19 | Llama 3.2 90B |  Meta | 52% | Epoch AI | ||
| 20 | Llama 4 Maverick | Meta | 52% | Epoch AI | ||
| 21 | Claude Opus 4 | Anthropic | 49% | 32K | Epoch AI | |
| 22 | Grok 4 | xAI | 45% | Epoch AI | ||
| 23 | Claude Sonnet 4 | Anthropic | 37% | Epoch AI | ||
| 24 | Pixtral 12B |  Mistral AI | 37% | Epoch AI | ||
| 25 | Claude 3.5 Haiku | Anthropic | 34% | Epoch AI |
Compare the leaders
Other multimodal benchmarks
- LMArena Vision
- Video-MME
- VPCT
- Blueprint-Bench 2
- Furniture Assembly
- LMArena Document (reference)
- MindCube (reference)
- ScienceQA (reference)
- SpatialViz-Bench (reference)
Frequently asked questions
What does GeoBench measure?
Guessing the country of a street-level photo, in the style of GeoGuessr.
Which model has the highest GeoBench score?
As of October 2026, Gemini 3 Flash Preview has the highest published GeoBench score on Noometry at 88%, out of 25 models with results.
What is the best open-weight model on GeoBench?
Qwen2.5-VL 72B Instruct has the highest GeoBench accuracy among open-weight models at 62%, ranking 17 of 25 overall.