RoboBench: leaderboard
Embodied-brain benchmark for multimodal LLMs across perception, instruction comprehension, planning, affordance prediction, and failure analysis.
Metric: Overall Dimension Average (imported). Source: robo-bench.github.io. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 50.1 | #152 |
| 2 | Gemini 2.5 Flash | 45.06 | #245 |
| 3 | Gemini 2.0 Flash | 45.04 | #342 |
| 4 | Claude 3.7 Sonnet | 40.53 | #249 |
| 5 | Qwen 2.5 VL 72B Instruct | 40.51 | #377 |
| 6 | GPT-4o | 40.16 | #343 |
| 7 | Claude 3.5 Sonnet | 37.82 | #350 |
| 8 | GPT-4o Mini | 34.4 | #599 |
| 9 | Qwen 2.5 VL 7B Instruct | 25.57 | #654 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=robobench · How It Works · Data refreshed daily, snapshot 2026-10-20.