ERGeoBench - Foundational Perception: leaderboard

Foundational Perception diagnostics on ERGeoBench (2026): multiple-choice and true or false questions about architecture, infrastructure, signage, terrain and vegetation in street-level views, averaged over the five categories. Accuracy in percent; higher is better.

Metric: Accuracy (%) on the foundational perception diagnostic questions: multiple-choice and true/false questions on architecture, infrastructure, signage, terrain and vegetation in a declared field of view, averaged over the five categories; unified structured prompt with egocentric observations, action history and in-context examples, temperature 0.1, first valid JSON object parsed (invalid outputs count as failures); higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-4o82.78#751
2Qwen-VL-Max81.1#1400
3InternVL3-8B75.99#1782
4Gemini 2.5 Pro70.26#276
5Qwen 3 VL 8B70.23#917
6InternVL2.5-8B69.45#2370
7Gemini 2.0 Flash69.08#723
8Gemini 3 Flash66.37#175
9Qwen 2.5 VL 7B61.52#1862

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=ergeobench-foundational-perception · How It Works · Data refreshed daily, snapshot 2026-11-21.