ImplicitQA (Average Accuracy): leaderboard
ImplicitQA video-reasoning benchmark (1,000 QA pairs over 320+ creative clips) testing human-like implicit reasoning — inferring motives, causality, and relationships across discontinuous frames rather than from explicit on-screen cues; reported as average accuracy.
Metric: Average Accuracy. Source: huggingface.co. Saturation forecast: Estimated already saturated. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 64.1 | #124 |
| 2 | GPT-4.1 | 54.3 | #248 |
| 3 | Qwen 2 VL 7B | 44.9 | #750 |
| 4 | Qwen 2.5 VL 7B | 42.8 | #643 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=implicitqa-average-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-20.