ReactHuman - Action-Intent Alignment: leaderboard
ReactHuman's Action-Intent Alignment: how well the model's chosen action serves the intent behind the human-like reaction, 0-1. Higher is better. Source: arxiv.org/abs/2609.10895.
Metric: AIA (0-1). Source: arxiv.org. Saturation forecast: Not forecast. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 3 27B | 0.61 | #606 |
| 2 | Qwen 3 VL 235B A22B | 0.48 | #251 |
| 3 | Gemini 2.5 Flash | 0.47 | #255 |
| 4 | GPT-5.5 | 0.47 | #33 |
| 5 | Claude Opus 4.8 | 0.41 | #39 |
| 6 | Kimi K2.6 | 0.32 | #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-action-intent-alignment · How It Works · Data refreshed daily, snapshot 2026-11-05.