EgoIntrospect - Valuable Interaction Identification: leaderboard
Metric: Accuracy (%) on 1,161 four-choice questions asking which of four proactive assistant requests the wearer accepted for a clip (distractors are requests the wearer did not accept), fixation-overlaid egocentric clips from 15 held-out participants with the participant profile; native audio-video models hear the clip audio, the other models read its transcript; temperature 0 (0.6 for InternVL3.5 thinking mode); higher is better. Source: arxiv.org. Saturation forecast: Around 2030. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 55.81 |
| 2 | Kimi K2.5 | 54.78 |
| 3 | Gemini 3 Flash (Preview) | 52.8 |
| 4 | Kimi K2.5 (Thinking) | 52.37 |
| 5 | Qwen 3.5 397B A17B | 51.42 |
| 6 | Qwen 3.5 Omni Plus | 49.96 |
| 7 | Qwen 3 VL 8B Instruct | 47.63 |
| 8 | Qwen3 Omni 30B A3B Instruct | 46.34 |
| 9 | Qwen 3 VL 8B (Thinking) | 46.17 |
| 10 | GPT-4o | 43.84 |
| 11 | InternVL3-8B | 42.89 |
| 12 | Qwen 2.5 VL 7B Instruct | 41.09 |
Interactive version: theaggregate.ai/benchmark?slug=egointrospect-valuable-interaction-identification · How It Works · Data refreshed daily, snapshot 2026-10-07.