IKEA-Bench - Video Discrimination: leaderboard
Metric: Accuracy (%) of video discrimination diagnostic (judge whether two video segments of one product show the same assembly step; 350 questions; chance 50) on IKEA-Bench, wordless IKEA manual diagrams against real assembly video from 29 products, Visual setting (diagram images only), zero-shot greedy decoding, unparsed answers counted wrong; higher is better. Source: arxiv.org. Saturation forecast: Around September 2027. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash | 71.1 |
| 2 | Gemini 3.1 Pro (Preview) | 67.4 |
| 3 | Qwen 3.5 27B | 63.7 |
| 4 | Qwen 3.5 9B | 63.1 |
| 5 | Gemma 3 27B | 53.7 |
| 6 | Qwen 3.5 2B | 51.4 |
| 7 | GLM-4.1V-9B (Thinking) | 50.3 |
| 8 | Gemma 3 12B | 49.1 |
| 9 | Gemma 3 4B | 47.7 |
| 10 | Qwen 2.5 VL 7B | 46 |
Interactive version: theaggregate.ai/benchmark?slug=ikea-bench-video-discrimination · How It Works · Data refreshed daily, snapshot 2026-10-07.