ReactHuman - Semantic Action Accuracy: leaderboard
Physics-grounded reactive decision-making for embodied multimodal models over 306 simulated scenes: whether the chosen action matches the human-like semantic action. Higher is better. Source: arxiv.org/abs/2609.10895.
Metric: SAA (%). Source: arxiv.org. Saturation forecast: Around 2031. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 63.7 | #33 |
| 2 | Gemma 3 27B | 62.4 | #606 |
| 3 | Qwen 3 VL 235B A22B | 56.4 | #251 |
| 4 | Claude Opus 4.8 | 50.7 | #39 |
| 5 | Gemini 2.5 Flash | 48.7 | #255 |
| 6 | Kimi K2.6 | 40.9 | #113 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=reacthuman-semantic-action-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.