KNOWS - Accessibility Tree + Screenshot: leaderboard
KNOWS partial credit (share of evaluation steps passed per task checkpoint, averaged) when a browser agent researches the live web and builds a Google Doc, Sheet or Slide deck, seeing accessibility trees and screenshots.
Metric: Average checkpoint fraction (%, overall, BrowserGym-AgentLab and Codex, accessibility tree and screenshots). Source: alexgill321.github.io. Saturation forecast: Around April 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 72.5 | #6 |
| 2 | GPT-5.5 | 49.5 | #33 |
| 3 | GPT-6 Luna | 48 | #61 |
| 4 | Claude Opus 4.7 | 42.5 | #51 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=knows-accessibility-tree-plus-screenshot · How It Works · Data refreshed daily, snapshot 2026-11-05.