EgoGapBench - Direct Actions: leaderboard
Metric: Accuracy (%; zero-shot multiple choice on 650 direct-action single-image items built from COCO multi-agent scenes: choose the action available from the given viewpoint, with distractors copying the singleton action of another visible agent (the correct answer is an active intervention); predicted option string-matched to the key; chance about 30%). Source: arxiv.org. Saturation forecast: Around March 2028. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Medium) | 50.9 |
| 2 | GPT-5.4 (Low) | 40.9 |
| 3 | gemma-4-E4B-it | 34 |
| 4 | Gemini 3 Flash (Preview) (Low) | 27.8 |
| 5 | Qwen 2.5 VL 7B Instruct | 21.8 |
| 6 | InternVL3.5-8B | 20.5 |
| 7 | Qwen 3.5 9B | 20 |
| 8 | GPT-4o (2024-11-20) | 12.9 |
Interactive version: theaggregate.ai/benchmark?slug=egogapbench-direct-actions · How It Works · Data refreshed daily, snapshot 2026-09-29.