EmbodiedBench: leaderboard
Benchmark for vision-driven embodied AI agents across navigation, manipulation, and household tasks (ALFRED, Habitat). ICML 2025 Oral.
Metric: Avg Score (%). Source: embodiedbench.github.io. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 67.7 | #249 |
| 2 | Claude 3.5 Sonnet | 64 | #350 |
| 3 | Gemini 1.5 Pro | 62.3 | #445 |
| 4 | GPT-4o | 56.3 | #343 |
| 5 | Gemini 2.0 Flash | 52.3 | #342 |
| 6 | Qwen 2.5 VL 72B Instruct | 39.7 | #377 |
| 7 | Gemini 1.5 Flash | 39.3 | #618 |
| 8 | InternVL3-78B | 39 | #352 |
| 9 | InternVL3-38B | 38 | #405 |
| 10 | InternVL2.5-78B | 37.7 | #431 |
| 11 | Gemma 3 27B (IT) | 37 | #523 |
| 12 | Llama 3.2 90B Vision Instruct | 32 | #622 |
| 13 | Gemma 3 12B (IT) | 25.7 | #664 |
| 14 | GPT-4o Mini | 24 | #599 |
| 15 | InternVL3-8B | 10.3 | #620 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=embodiedbench · How It Works · Data refreshed daily, snapshot 2026-10-20.