Reasoning
Furniture Assembly Leaderboard
Furniture Assembly, built by Epoch AI, tests spatial reasoning with 60 photos of partly built IKEA furniture and the assembly manual: the model must say whether the build is correct so far and, if not, name the mistaken step and describe the error. Scores are accuracy across the photos, and a human baseline averaged 63%.
Source: epoch3 open models ranked+31 proprietaryData through Oct 2026
All models ranked on Furniture Assembly
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | Claude Opus 5.5 Max · proprietary | 83.3% | API only |
| 2 | GPT 6 Astra Max · proprietary | 80.0% | API only |
| 3 | GPT 6.1 Sol Max · proprietary | 80.0% | API only |
| 4 | Claude Sonnet 5.5 Max · proprietary | 75.0% | API only |
| 5 | Gemini 4 Argon (high) · proprietary | 72.5% | API only |
| 6 | Claude Fable 5.1 Max · proprietary | 70.0% | API only |
| 7 | Claude Opus 5 Max · proprietary | 60.8% | API only |
| 8 | GPT 6 Sol Max · proprietary | 58.3% | API only |
| 9 | GPT 5.6 Sol Max · proprietary | 56.7% | API only |
| 10 | GPT 5.6 Terra Max · proprietary | 54.2% | API only |
| 11 | Muse Wonder 1.0 Max · proprietary | 54.2% | API only |
| 12 | Claude Haiku 5.5 Max · proprietary | 47.5% | API only |
| 13 | GPT 5.5 (xhigh) · proprietary | 44.2% | API only |
| 14 | GPT 6 Luna Max · proprietary | 44.2% | API only |
| 15 | Claude Opus 4.8 Max · proprietary | 42.5% | API only |
| 16 | GPT 5.6 Luna Max · proprietary | 42.5% | API only |
| 17 | Grok 4.6 (xhigh) · proprietary | 40.0% | API only |
| 18 | GPT 5.2 (Dec 11, 2025, xhigh) · proprietary | 38.3% | API only |
| 19 | GPT 5.4 (Mar 05, 2026, xhigh) · proprietary | 37.5% | API only |
| 20 | Claude Fable 5 Max · proprietary | 35.8% | API only |
| 21 | DeepSeek V4.1 Flash · 763.2B | 34.2% | ~458 GB |
| 22 | Kimi K3 · 2779.9B | 34.2% | ~1674 GB |
| 23 | Claude Opus 4.7 Max · proprietary | 33.3% | API only |
| 24 | Gemini 3.8 Flash (high) · proprietary | 31.7% | API only |
| 25 | Claude Opus 4.5 (Nov 01, 2025, 64K) · proprietary | 28.3% | API only |
| 26 | Claude Opus 4.6 Max · proprietary | 28.3% | API only |
| 27 | Gemini 3.1 Pro Preview (high) · proprietary | 26.7% | API only |
| 28 | Gemini 3.7 Flash (high) · proprietary | 26.7% | API only |
| 29 | Gemini 3.6 Flash (high) · proprietary | 23.3% | API only |
| 30 | Grok 4.5 (high) · proprietary | 22.5% | API only |
| 31 | Kimi K2.6 · 1026.9B | 21.7% | ~620 GB |
| 32 | Grok 4.7 (xhigh) · proprietary | 20.8% | API only |
| 33 | Mistral Large 4.0 (high) · proprietary | 20.8% | API only |
| 34 | Qwen3.8 Max (Sep 02, xhigh) · proprietary | 20.0% | API only |
Furniture Assembly: frequently asked questions
- What is the best open LLM on Furniture Assembly?
- DeepSeek V4.1 Flash is the top open model on Furniture Assembly, scoring 34.2%. Among all models tested — including proprietary ones — it ranks #21. The top model overall is Claude Opus 5.5 Max (Anthropic) at 83.3%.
- Can open models match proprietary models on Furniture Assembly?
- Not quite on Furniture Assembly: the strongest proprietary model (Claude Opus 5.5 Max) scores 83.3%, ahead of the best open model (DeepSeek V4.1 Flash) at 34.2% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.