GeoR-Bench - Reasoning: leaderboard
Metric: Reasoning score (0-100): rubric-weighted judgment of whether the edit realises the intended physical, spatial or process-based change, compared with the ground-truth output (mean of the six category scores), 440 reasoning-informed geoscience image-editing samples in six categories (geomorphology, hydrology, atmosphere and ocean, cryosphere, GIS and spatial geometry, crustal science), one generated output per sample, Gemini 3 Flash judge with task-specific rubrics; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-Image-2 | 79.1 |
| 2 | Nano Banana 2 | 76.2 |
| 3 | Nano Banana Pro | 71.4 |
| 4 | GPT-Image-1.5 | 68.1 |
| 5 | Seedream 5.0 | 59.2 |
| 6 | Nano Banana | 54 |
| 7 | Qwen-Image-Edit-2511 | 48.9 |
| 8 | FLUX 2 Max | 48.5 |
| 9 | Seedream 4.5 | 46.5 |
| 10 | FLUX 2 Pro | 42.6 |
Interactive version: theaggregate.ai/benchmark?slug=geor-bench-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-07.