GapEval - Numerical Perception (Understanding): leaderboard
Metric: Accuracy (%; count two object types in an image and state the swapped counts; text answers to GapEval's bidirectional questions, judged correct or incorrect by GPT-5-mini against reference answers, mean of ten sampled runs). Source: arxiv.org. Saturation forecast: Around 2030. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 Mini | 30.3 |
| 2 | Qwen 3 VL 235B A22B | 26 |
| 3 | Qwen 3 VL 8B | 25 |
| 4 | GPT-5 | 24.05 |
| 5 | Gemini 2.5 Flash | 17 |
Interactive version: theaggregate.ai/benchmark?slug=gapeval-numerical-perception-understanding · How It Works · Data refreshed daily, snapshot 2026-09-26.