GeoBench leaderboard
- models measured
- 12
- lead over #2
- +7.0 pts
- median
- 71.5%
Leader on GeoBench · read October 22
On October 22, 2026, 18 of our catalog's models had a GeoBench result, and Google Gemini 3 Flash Preview led with 88.0%, ahead of OpenAI GPT-5 at 81.0%.
Locating a place from a photo, as in GeoGuessr. Share of country guesses correct.
Role in the index
GeoBench is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | Gemini 3 Flash PreviewGoogle | 88.0% | 43.7 | $3.00 | |
| 2 | GPT-5OpenAI | 81.0% | 38.8 | $10.00 | |
| 3 | o1OpenAI | 80.0% | Not benchmarked | $60.00 | |
| 4 | Gemini 2.0 FlashRetiredGoogle | 77.0% | Not benchmarked | $0.40 | |
| 5 | Gemini 1.5 FlashRetiredGoogle | 76.0% | Not benchmarked | $0.30 | |
| 6 | Claude Opus 4.5Anthropic | 75.0% | 44.0 | $25.00 | |
| 7 | o3OpenAI | 74.0% | Not benchmarked | $8.00 | |
| 8 | GPT-4.1OpenAI | 72.0% | Not benchmarked | $8.00 | |
| 9 | GPT-4o (2024-11-20)OpenAI | 71.0% | Not benchmarked | $10.00 | |
| 10 | Claude 3.7 SonnetRetiredAnthropic | 68.0% | Not benchmarked | $15.00 | |
| 11 | GPT-4o-mini (2024-07-18)OpenAI | 64.0% | Not benchmarked | $0.60 | |
| 12 | o4 MiniOpenAI | 64.0% | Not benchmarked | $4.40 | |
| 13 | Claude 3.5 SonnetRetiredAnthropic | 62.0% | Not benchmarked | $15.00 | |
| 14 | Qwen2.5 VL 72B InstructQwen | 62.0% | Not benchmarked | $1.00 | |
| 15 | Gemma 3 27BGoogle | 52.0% | Not benchmarked | $0.30 | |
| 16 | Llama 4 MaverickMeta Llama | 52.0% | Not benchmarked | $1.00 | |
| 17 | Claude Opus 4RetiredAnthropic | 49.0% | Not benchmarked | $75.00 | |
| 18 | Claude 3.5 HaikuRetiredAnthropic | 34.0% | Not benchmarked | $4.00 |
Only models in our catalog are listed, and each appears once with its best published run. 215 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out.