Act2Intention - Intention Prediction Category Accuracy: leaderboard
Act2Intention Bench (arXiv 2608.14132v1, Table 7): mobile GUI action trajectories, in-distribution test set, general models without task fine-tuning. Accuracy of the predicted category of the user's next intention from intention history and persona, percent. Higher is better.
Metric: Correct next-intention category (%). Source: arxiv.org. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 7B | 43.5 | #895 |
| 2 | GPT-3.5 Turbo | 42.5 | #858 |
| 3 | Qwen 2.5 72B | 40 | #512 |
| 4 | GPT-4o Mini | 38.5 | #598 |
| 5 | Llama 3.1 8B | 37.8 | #1130 |
| 6 | Mistral 7B | 35.34 | #1382 |
| 7 | GPT-4o | 34.5 | #357 |
| 8 | Gemini 1.5 Pro | 34.5 | #454 |
| 9 | Claude 3.5 Haiku | 28.5 | #571 |
| 10 | Claude 3.5 Sonnet | 15.5 | #362 |
Interactive version: theaggregate.ai/benchmark?slug=act2intention-intention-prediction-category-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.