WebRetriever (Protocol II, With Documentation): leaderboard
Metric: Success rate (%; 1,000 navigation tasks, 550 shared with Protocol I, with operational documentation for the site; live websites from WebRetriever (800 sites across consumer, professional and enterprise domains), each task starting from a fixed entry URL with search engines blocked, Playwright-based online harness; task success judged by human annotators). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Pro (Harness: Computer-use) | 45.2 |
| 2 | Claude Sonnet 4.5 (Harness: Computer-use) | 36.3 |
| 3 | Browser Use + GPT-4o (2024-08-06) | 31.6 |
| 4 | UI-TARS-1.5 | 24.8 |
| 5 | Agent-E + GPT-4o (2024-08-06) | 20.4 |
| 6 | SeeAct + GPT-4o (2024-08-06) | 17.1 |
Interactive version: theaggregate.ai/benchmark?slug=webretriever-protocol-ii-with-documentation · How It Works · Data refreshed daily, snapshot 2026-09-29.