LiveWeb-IE (Reflexion): leaderboard
Metric: Overall F1 (%) of extracted values over all task types of LiveWeb-IE (live websites in 15 domains): the model writes reusable XPath wrappers from the first page of each page group, given its HTML and a natural-language query, and the wrappers extract values from every page of the group on the live site; attributes are aligned by a GPT-4o matcher; Reflexion, which refines the wrapper from execution failures; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 24.22 | #333 |
| 2 | Gemini 2.5 Flash | 22.39 | #237 |
| 3 | Qwen 2.5 32B Instruct | 20.28 | #491 |
| 4 | Qwen 2.5 72B Instruct | 18.36 | #436 |
| 5 | Gemma 3 27B (IT) | 17.47 | #509 |
| 6 | GPT-4o Mini | 17.47 | #588 |
| 7 | Qwen 2.5 7B Instruct | 14.53 | #846 |
| 8 | Gemma 3 4B (IT) | 10.35 | #971 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=liveweb-ie-reflexion · How It Works · Data refreshed daily, snapshot 2026-10-11.