ProEvent - Single-Step Recall: leaderboard
ProEvent (2026) tests proactive agents that keep a user timetable current from ongoing instant-messaging chats with concurrent threads and noise, inserting, updating or deleting events as plans change. Recall of the ground-truth timetable operations over 2,463 single-step questions, in percent; higher is better.
Metric: Recall of ground-truth operations (%; over the 2,463 single-step questions, insert, update and delete pooled; synthesized instant-messaging chats (1,249 chats, concurrent threads, injected noise) from which a proactive agent must keep the user timetable up to date by inserting, updating or deleting events at each time window). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 58 | #229 |
| 2 | DeepSeek R1 | 56.7 | #473 |
| 3 | DeepSeek V3.2 | 46.1 | #406 |
| 4 | Qwen 3 235B A22B | 37.3 | #623 |
| 5 | Qwen 3 8B | 27.5 | #1830 |
| 6 | Qwen 3 32B | 19.8 | #960 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=proevent-single-step-recall · How It Works · Data refreshed daily, snapshot 2026-11-13.