BrowserGym WebLINX: leaderboard
WebLINX's recorded human web-navigation dialogues replayed on ServiceNow's BrowserGym leaderboard as single-step tasks: the agent predicts the next action, which is matched against the recorded human action. Scored with the dataset's turn-level partial-match metrics (%); higher is better. Runs use AgentLab's GenericAgent scaffold unless the row names another framework.
Metric: Turn-level score (%, GenericAgent unless named). Source: huggingface.co. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 13.7 | #635 |
| 2 | GPT-4o (2024-08-06) | 12.5 | #731 |
| 3 | O1 Mini (2024-09-12) | 12.5 | #794 |
| 4 | GPT-4o Mini (2024-07-18) | 11.6 | #1512 |
| 5 | Llama 3.1 70B | 8.9 | #1493 |
| 6 | Llama 3.1 405B | 7.9 | #1083 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=browsergym-weblinx · How It Works · Data refreshed daily, snapshot 2026-11-12.