EarthSpatialBench (Direct) - Within Counting: leaderboard
Metric: Mean absolute error of the object count on EarthSpatialBench containment (within) counting questions on 512x512 SatlasPretrain Earth images at about 1 m resolution, direct prompting; lower is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 30B A3B Instruct | 2.45 | #365 |
| 2 | GPT-5 | 2.6 | #91 |
| 3 | Qwen 3 VL 30B A3B (Thinking) | 2.79 | #338 (Qwen 3 VL 30B A3B) |
| 4 | Claude 3.5 Sonnet | 2.83 | #337 |
| 5 | Gemini 2.5 Pro | 2.94 | #145 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=earthspatialbench-direct-within-counting · How It Works · Data refreshed daily, snapshot 2026-10-11.