Vals AI Robo-bench: leaderboard
Vals AI evaluates models controlling a robot arm and humanoid in real-time physics simulation from motor commands and camera images. Accuracy measures success on robot tasks, most of which are held out; this is an agent-control protocol, not static visual question answering.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around 2029. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6.1 Sol (Max) | 28.33 | #8 (GPT-6.1 Sol) |
| 2 | GPT-6 Astra (Max) | 23.33 | #5 (GPT-6 Astra) |
| 3 | Claude Opus 5.5 (Max) | 14.58 | #3 (Claude Opus 5.5) |
| 4 | Muse Wonder Full (Max) | 10.42 | #11 (Muse Wonder Full) |
| 5 | Claude Fable 5.1 (Max) | 5.83 | #9 (Claude Fable 5.1) |
| 6 | Gemini 4 Argon (High) | 0.83 | #7 (Gemini 4 Argon) |
| 7 | Grok 4.7 (xHigh) | 0.21 | #27 (Grok 4.7) |
No result here: #10 Claude Sonnet 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-robo-bench · How It Works · Data refreshed daily, snapshot 2026-10-16.