GRAB-Lite — leaderboard
GRaph Analysis Benchmark: 500 task-balanced questions testing multimodal models on graph/figure analysis — estimating means, intercepts, correlations, and transforms across synthetic and realistic figures.
Metric: Overall Score. Source: grab-benchmark.github.io. Status: saturation imminent. 38 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 | 74 |
| 2 | GPT-5.5 | 71.8 |
| 3 | GPT-5.4 | 71 |
| 4 | Gemini 3.5 Flash | 63 |
| 5 | GPT-5.4 Mini | 62 |
| 6 | Claude Opus 4.8 | 60.6 |
| 7 | Gemini 3.1 Pro (Preview) | 59.4 |
| 8 | GPT-5.2 | 59 |
| 9 | Claude Opus 4.7 | 58.2 |
| 10 | Gemini 3 Flash | 58.2 |
| 11 | Gemini 3 Pro | 53.4 |
| 12 | Gemini 3.1 Flash Lite | 47.8 |
| 13 | Claude Sonnet 4.6 | 46.6 |
| 14 | GPT-5 | 46.2 |
| 15 | Claude Opus 4.6 | 45.6 |
Interactive version: theaggregate.ai/benchmark?slug=grab-lite · How the rankings work · Data refreshed daily, snapshot 2026-07-22.