MAG (Coordinates) - Gated Guide Score: leaderboard
The gated guide score of MAG (2026) with raw pixel-coordinate actions: the quality of the step-by-step guide a web agent writes, counted only on tasks it completes and averaged over the 171 test tasks with reference guides, in percent. Higher is better.
Metric: Gated guide score (%): zero on a failed task and 0.4 + 0.6 times the guide-overlap score on a solved one, averaged over the 171 test tasks with reference guides, raw pixel coordinates grounding; higher is better. Source: arxiv.org. Saturation forecast: Around January 2028. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (Low) | 22.5 | #49 (GPT-5.5) |
| 2 | Claude Sonnet 4.6 (Low) | 14.5 | #161 (Claude Sonnet 4.6) |
| 3 | Gemini 3.5 Flash (Low) | 10.3 | #101 (Gemini 3.5 Flash) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mag-coordinates-gated-guide-score · How It Works · Data refreshed daily, snapshot 2026-11-13.