IKEA-Bench - Step Verification: leaderboard

Metric: Accuracy (%) of binary step verification (judge whether a video segment matches a given manual diagram or an adjacent step's diagram; 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 2028. 19 models tracked.

Top models

#ModelScore
1Gemini 3 Flash68.6
2Gemini 3.1 Pro (Preview)65.1
3Qwen 3.5 9B63.7
4Qwen 3.5 27B62.9
5Qwen 3.5 2B56.6
6Gemma 3 27B55.7
7GLM-4.1V-9B (Thinking)55.7
8Qwen 2.5 VL 7B50.9
9Gemma 3 4B50.3
10Gemma 3 12B49.7

Interactive version: theaggregate.ai/benchmark?slug=ikea-bench-step-verification · How It Works · Data refreshed daily, snapshot 2026-10-07.