Act2Intention - Intention Understanding Group Accuracy: leaderboard
Act2Intention Bench (arXiv 2608.14132v1, Table 7): mobile GUI action trajectories, in-distribution test set, general models without task fine-tuning. Share of action sequences segmented exactly into the reference intention units, percent. Higher is better.
Metric: Correct intention segmentation (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Pro | 68.63 | #454 |
| 2 | Claude 3.5 Haiku | 24.28 | #571 |
| 3 | GPT-4o Mini | 20.66 | #598 |
| 4 | GPT-4o | 20.08 | #357 |
| 5 | Claude 3.5 Sonnet | 20 | #362 |
| 6 | GPT-3.5 Turbo | 16.31 | #858 |
| 7 | Qwen 2.5 7B | 12.14 | #895 |
| 8 | Llama 3.1 8B | 6.03 | #1130 |
| 9 | Mistral 7B | 5.63 | #1382 |
| 10 | Qwen 2.5 72B | 5.33 | #512 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=act2intention-intention-understanding-group-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.