MobilePA-Bench - Memory Usage: leaderboard
End-to-end success on MobilePA-Bench tasks that require recalling stored memories, user profiles and past preferences to resolve implicit requests, verified by database state or behavior (MobilePA-Bench, arXiv 2608.23035v2, Table 3). Percent; higher is better.
Metric: End-to-end success (%). Source: arxiv.org. Saturation forecast: Around May 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.8 Max | 64.63 | #34 |
| 2 | Kimi K3 | 63.56 | #35 |
| 3 | Gemini 3.6 Flash | 62.77 | #57 |
| 4 | Claude Fable 5 | 62.5 | #20 |
| 5 | Claude Opus 5 | 58.51 | #16 |
| 6 | Qwen 3.7 Max | 53.19 | #85 |
| 7 | GLM-5.2 | 49.73 | #80 |
| 8 | Gemini 3.1 Pro (Preview) | 48.67 | #64 |
| 9 | GPT-5.6 Sol | 44.15 | #22 |
| 10 | Seed 2.1 Pro | 42.29 | #73 |
| 11 | GPT-5.5 | 41.76 | #33 |
| 12 | Claude Opus 4.8 | 37.23 | #39 |
| 13 | Kimi K2.6 | 33.78 | #113 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mobilepa-bench-memory-usage · How It Works · Data refreshed daily, snapshot 2026-11-05.