WebRetriever (Protocol III, End-to-End): leaderboard
WebRetriever (2026) Protocol III: 100 end-to-end web tasks that require navigating live websites and extracting a unique factual answer from text, documents or charts. Human-judged success rate in percent, higher is better.
Metric: Success rate (%; 100 end-to-end tasks that require navigating and extracting a unique factual answer from text, documents or charts; 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: Around January 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 11 | #740 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=webretriever-protocol-iii-end-to-end · How It Works · Data refreshed daily, snapshot 2026-11-13.