Atlas

Benchmarks

← All benchmarks

GeoBench ACW Country Accuracy

Multimodal · 2025-03-22

GeoBench ACW Country Accuracy is the percentage of the 100 ACW (A Community World) locations for which the model names the correct country, from the same runs as the GeoBench headline score. Refused rounds count as misses.

Top models (higher is better)

ModelScore
Claude Fable 589.0
Gemini 3 Flash Preview88.0
Gemini 2.5 Pro Preview 05-0686.0
Gemini 3.8 Flash86.0
Muse Spark 1.386.0
Gemini 3.5 Flash85.0
Claude Fable 5.184.0
Gemini 3.1 Pro Preview84.0
Gemini 3 Pro Preview84.0
Gemini 3.1 Flash-Lite Preview83.0
Opus 4.881.0
GPT-581.0
Gemini 2.5 Pro Experimental 03-2578.0
Gemini 2.0 Flash77.0
Gemini 1.5 Flash (source unspecified)76.0
Gemini 2.5 Flash76.0
Opus 4.575.0
Sonnet 574.0
o374.0
Sonnet 4.673.0
Gemini 2.5 Flash Preview 04-1773.0
GPT-5.6 Sol73.0
GPT-4.172.0
GLM-4.6V71.0
GPT-4o (source unspecified)71.0
GPT-4o Mini64.0
O4 Mini64.0
Claude 3.5 Sonnet (Oct 2024)62.0
Qwen2.5-VL-72B-Instruct62.0
Grok 4.761.0
Sonnet 4.560.0
Qwen3-VL-235B-A22B-Instruct59.0
Grok 4.356.0
Mistral Medium 3.556.0
Gemma 3 27B IT52.0
Llama 4 Maverick Instruct52.0
Opus 449.0
Grok 445.0
Qwen3.8 27B38.0
Sonnet 437.0
Loading Atlas data…