Benchmarks · reasoning
Furniture Assembly
Epoch AI's spatial-reasoning benchmark: from photos of IKEA builds (Ställ shoe cabinet, Tonstad bed, Gullaberg dresser) and the official manual, the model decides whether the build is correct and, if not, identifies and describes the mistaken step. Mean score. Data: Epoch AI, CC BY.
34 models · 34 results · data as of 2026-10-29 · epoch.ai ↗
Leaderboard
Opus 5.5 max
All results
34 rows| # | Model | Effort | Furniture Assembly | Harness |
|---|---|---|---|---|
| 1 | Opus 5.5 | max | 83.3% | Epoch AI |
| 2 | GPT-6 Astra | max | 80.0% | Epoch AI |
| 3 | GPT-6.1 Sol | max | 80.0% | Epoch AI |
| 4 | GPT-6.1 Terra | max | 79.2% | Epoch AI |
| 5 | Sonnet 5.5 | max | 75.0% | Epoch AI |
| 6 | Gemini 4 Argon | high | 72.5% | Epoch AI |
| 7 | Fable 5.1 | max | 70.0% | Epoch AI |
| 8 | Opus 5 | max | 60.8% | Epoch AI |
| 9 | GPT-6 Sol | max | 58.3% | Epoch AI |
| 10 | GPT-5.6 Sol | max | 56.7% | Epoch AI |
| 11 | GPT-5.6 Terra | max | 54.2% | Epoch AI |
| 12 | Haiku 5.5 | max | 47.5% | Epoch AI |
| 13 | GPT-5.5 | xhigh | 44.2% | Epoch AI |
| 14 | GPT-6 Luna | max | 44.2% | Epoch AI |
| 15 | Opus 4.8 | max | 42.5% | Epoch AI |
| 16 | GPT-5.6 Luna | max | 42.5% | Epoch AI |
| 17 | Grok 4.6 | xhigh | 40.0% | Epoch AI |
| 18 | GPT-5.2 | xhigh | 38.3% | Epoch AI |
| 19 | GPT-5.4 | xhigh | 37.5% | Epoch AI |
| 20 | Fable 5 | max | 35.8% | Epoch AI |
| 21 | DeepSeek V4.1 Flash | max | 34.2% | Epoch AI |
| 22 | Kimi K3 | max | 34.2% | Epoch AI |
| 23 | Opus 4.7 | max | 33.3% | Epoch AI |
| 24 | Gemini 3.8 Flash | high | 31.7% | Epoch AI |
| 25 | Opus 4.5 | high | 28.3% | Epoch AI |
| 26 | Opus 4.6 | max | 28.3% | Epoch AI |
| 27 | Gemini 3.1 Pro | high | 26.7% | Epoch AI |
| 28 | Gemini 3.7 Flash | high | 26.7% | Epoch AI |
| 29 | Gemini 3.6 Flash | high | 23.3% | Epoch AI |
| 30 | Grok 4.5 | high | 22.5% | Epoch AI |