Butter-Bench: leaderboard
Butter-Bench (Andon Labs, 2025): an LLM orchestrates a robot vacuum with lidar and a camera through high-level actions on six household delivery tasks, from finding the package with the butter to a full end-to-end delivery. Score is the average completion rate over the six tasks (5 trials each); higher is better, 0-100 (%). Humans averaged 95%.
Metric: Average completion rate (%). Source: andonlabs.com. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 40 | #159 |
| 2 | Claude Opus 4.1 | 37 | #146 |
| 3 | GPT-5 | 30 | #105 |
| 4 | Grok 4 | 23 | #189 |
| 5 | Llama 4 Maverick | 7 | #486 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=butter-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.