SentinelBench (wait_for Tool): leaderboard
SentinelBench (Microsoft Research, 2026) evaluates long-running monitoring agents on 100 tasks in synthetic web apps where scripted events unfold over time and the agent must act when a condition is met, or correctly do nothing. This board is task success rate in percent for a screenshot-based browser agent given a wait_for tool that blocks until the page meets a condition; higher is better.
Metric: Task success rate (%) over 100 time-evolving monitoring tasks in 10 synthetic web apps (no-op, passive and active tasks, 10-minute scripted event playback); screenshot-based browser agent adapted from Magentic-UI, with the purpose-built wait_for(condition, timeout) tool that blocks until an LLM-checked page diff meets the condition. Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (Low) | 75 | #123 (GPT-5.4) |
| 2 | GPT-4o | 48 | #741 |
| 3 | Qwen 3.5 9B | 48 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sentinelbench-wait-for-tool · How It Works · Data refreshed daily, snapshot 2026-11-13.