Epoch AI - Furniture Assembly: leaderboard
Epoch AI's Furniture Assembly Benchmark for spatial reasoning about physical construction: given one of 60 photos taken while building three IKEA products (Ställ shoe cabinet, Tonstad bed frame, Gullaberg dresser), the official assembly manual PDF, a zoom tool and a Python interpreter, the model must say whether the build is correct so far and, if not, name the step where each deliberate mistake was made and describe it. Many photos continue past the mistake so it is not the latest step. Higher is better.
Metric: Score. Source: epoch.ai. Saturation forecast: Around December 2027. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Max) | 83.33 | #3 (Claude Opus 5.5) |
| 2 | GPT-6 Astra (Max) | 80 | #5 (GPT-6 Astra) |
| 3 | GPT-6.1 Sol (Max) | 80 | #6 (GPT-6.1 Sol) |
| 4 | GPT-6.1 Terra (Max) | 79.17 | #14 (GPT-6.1 Terra) |
| 5 | Claude Sonnet 5.5 (Max) | 75 | #13 (Claude Sonnet 5.5) |
| 6 | Gemini 4 Argon (High) | 72.5 | #9 (Gemini 4 Argon) |
| 7 | Claude Fable 5.1 (Max) | 70 | #10 (Claude Fable 5.1) |
| 8 | Claude Opus 5 (Max) | 60.83 | #15 (Claude Opus 5) |
| 9 | GPT-6 Sol (Max) | 58.33 | #17 (GPT-6 Sol) |
| 10 | GPT-5.6 Sol (Max) | 56.67 | #22 (GPT-5.6 Sol) |
| 11 | Kimi K3.1 Pro (Max) | 55.83 | #18 (Kimi K3.1 Pro) |
| 12 | GPT-5.6 Terra (Max) | 54.17 | #43 (GPT-5.6 Terra) |
| 13 | Muse Wonder (Max) | 54.17 | #16 (Muse Wonder) |
| 14 | Claude Haiku 5.5 (Max) | 47.5 | #46 (Claude Haiku 5.5) |
| 15 | GPT-5.5 (xHigh) | 44.17 | #34 (GPT-5.5) |
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-furniture-assembly · How It Works · Data refreshed daily, snapshot 2026-11-07.