Memory-World - History-Retention Protocol: leaderboard
Memory-intensive mobile GUI tasks from Memory-World (STAMP paper, Wang et al., 2026): in synthesized virtual apps, information shown on one screen leaves the visual context and must be recalled later, and the environment verifies whether the task was completed. This board covers GUI-specialized agents run with their native prompts while the full dialogue history is kept through the rollout; the score is task accuracy on one attempt (pass@1), in percent; higher is better.
Metric: Task Accuracy (%; pass@1, GUI-specialized agents, native prompts with full dialogue history). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GUI-Owl-1.5-8B-Think | 44.9 | |
| 2 | MAI-UI-8B | 40.2 | |
| 3 | UI-TARS-1.5-7B | 20.5 | |
| 4 | CogAgent-9B-20241220 | 19.7 | |
| 5 | UI-Venus-1.5-8B | 18.9 |
Interactive version: theaggregate.ai/benchmark?slug=memory-world-history-retention-protocol · How It Works · Data refreshed daily, snapshot 2026-11-12.