AirGroundBench - Counterfactual Reasoning: leaderboard
Metric: Accuracy (%) on the counterfactual spatial-reasoning VQA task; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 Plus | 48.57 |
| 2 | Qwen 3.5 397B A17B | 45.71 |
| 3 | GPT-5.2 | 44.93 |
| 4 | Seed 2.0 Pro | 44.62 |
| 5 | Claude Sonnet 4.6 | 44.35 |
| 6 | Gemini 3.1 Pro (Preview) | 43.42 |
| 7 | Qwen 3.5 27B | 43.25 |
| 8 | Kimi K2.5 | 40.3 |
| 9 | GLM-4.6V | 38.46 |
| 10 | ERNIE 5.0 | 35.82 |
| 11 | Qwen 3 VL 235B A22B | 35.38 |
Interactive version: theaggregate.ai/benchmark?slug=airgroundbench-counterfactual-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-29.