SuperMemory-VQA (Video-RAG): leaderboard
SuperMemory-VQA is an egocentric long-horizon memory question-answering benchmark of 4,853 QA pairs from 52.9 hours of Meta Aria glasses recordings across six memory tasks, with ordered answer choices (correct, vague, wrong, unanswerable). This board reports Four-way QA accuracy (%) under the Video-RAG retrieval framework; only the correct option is credited over the four ordered choices; higher is better.
Metric: Four-way QA accuracy (%) under the Video-RAG retrieval framework; only the correct option is credited over the four ordered choices; higher is better. Source: arxiv.org. Saturation forecast: Around May 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 61 | #164 |
| 2 | Gemini 3.1 Pro (Preview) | 53.2 | #94 |
| 3 | GPT-5.4 | 52.3 | #123 |
| 4 | GPT-5.4 Mini | 47.8 | #389 |
| 5 | Gemma 4 31B (IT) | 45.6 | #342 |
| 6 | gemma-4-E4B-it | 35.3 | #1469 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=supermemory-vqa-video-rag · How It Works · Data refreshed daily, snapshot 2026-11-13.