GeoBench
Higher is better · Image-input models only
Like the game GeoGuessr: from a single photo, use clues such as signs, vegetation, and buildings to work out where it was taken. The value here is the share of photos on the 'A Common World' map where the country was guessed correctly. Scores run from 0 to 100%. Higher is better.
Top score88.0%Gemini 3 Flash
Models tested7
Model release date (newest on the right)
Best score so farHigher on the chart is better
Which AI gives the photo and chart answers people prefer most?Arena Vision
Claude Fable 5🥇Claude Fable 5🥇Can it spot the wrong step from assembly photos?Furniture AssemblyClaude Opus 5.5🥇Claude Opus 5.5🥇Can it draw a floor plan from a few photos of a home?Blueprint-Bench 2
Gemini 4 Argon🥇Gemini 4 Argon🥇
Claude Fable 5🥇Claude Fable 5🥇Can it spot the wrong step from assembly photos?Furniture AssemblyClaude Opus 5.5🥇Claude Opus 5.5🥇Can it draw a floor plan from a few photos of a home?Blueprint-Bench 2Gemini 4 Argon🥇Gemini 4 Argon🥇- 🥇GoogleGemini 3 Flash88.0%
- 🥈OpenAIGPT-5Reasoning effort: Medium81.0%
- 🥉AnthropicClaude Opus 4.575.0%
- #4OpenAIGPT-4.172.0%
- #5MetaLlama 4 Maverick52.0%
- #6AnthropicClaude Opus 4Reasoning effort: Budget 32K49.0%
- #7AnthropicClaude Sonnet 437.0%
Half of models ≤ 72.0%Last updated 2026-10-07
The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.