EmbodiedBench Navigation: leaderboard
Low-level embodied control in AI2-THOR, where the model emits discrete movement actions to locate and reach a target object across base, commonsense, complex-instruction, visual, and long-horizon splits.
Metric: Avg Score (%). Source: embodiedbench.github.io. Saturation forecast: Estimated already saturated. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 57.7 | #343 |
| 2 | InternVL3-78B | 53.7 | #352 |
| 3 | InternVL3-38B | 50.7 | #405 |
| 4 | Gemini 2.0 Flash | 48.7 | #342 |
| 5 | Gemma 3 27B (IT) | 45.4 | #523 |
| 6 | Claude 3.7 Sonnet | 45 | #249 |
| 7 | Claude 3.5 Sonnet | 44.7 | #350 |
| 8 | Gemini 1.5 Flash | 41.7 | #618 |
| 9 | Qwen 2.5 VL 72B Instruct | 40 | #377 |
| 10 | Gemma 3 12B (IT) | 34 | #664 |
| 11 | GPT-4o Mini | 32.8 | #599 |
| 12 | InternVL2.5-78B | 30.7 | #431 |
| 13 | Llama 3.2 90B Vision Instruct | 30 | #622 |
| 14 | InternVL3-8B | 29.3 | #620 |
| 15 | Gemini 1.5 Pro | 24.3 | #445 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=embodiedbench-navigation · How It Works · Data refreshed daily, snapshot 2026-10-20.