GeoR-Bench - Reasoning: leaderboard

Reasoning score on a 0-100 scale: whether an edit realizes the intended physical, spatial or process change compared with the ground-truth output, averaged over 440 geoscience image-editing samples in six categories (GeoR-Bench, 2026), judged by Gemini 3 Flash with task-specific rubrics. Higher is better.

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. 19 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT Image 279.1
2Nano Banana 276.2
3Nano Banana Pro71.4
4GPT Image 1.568.1
5Seedream 5.059.2
6Nano Banana54
7Qwen-Image-Edit-251148.9
8FLUX 2 Max48.5
9Seedream 4.546.5
10FLUX 2 Pro42.6
11Seedream 4.038.7
12FLUX.2-dev33.6
13Step1X-Edit24.7
14DreamOmni23.3
15Step1X-Edit (Thinking + Reflection)21.3

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