Summary statistics
Leaderboard
↑ Top| Model | Source | Status | Run ID |
|---|
WCAG Skill
↑ TopBoost your model's score with the wcag-skill v1.1.0 — an agent skill distilled from these benchmarks. Available on ClawHub.
The skill includes a fast-path template (known-good page with 107 WHATWG tags), a tag-coverage pre-check script, iteration logs from passing runs (DeepSeek V4 Pro: SCORE 191.8, max), and a proven 5-step workflow: Audit → Triage → Fix → Verify → Prevent. Skip the wasted first iteration — copy the template, replace placeholders, and start scoring on iteration 1.
Run History
↑ TopTimeline
↑ TopScore evolution over time — all models
Each point is a benchmark run plotted by date. Shows how models improve across re-tests.
Model Strategies
↑ TopScore progression — all models compared
Average score at each iteration across all runs per model
Efficiency — iterations vs final score
Each dot is a run. Top-left = best strategy (high score, few iterations)
Score velocity — points gained per iteration
Average score improvement per iteration (best run per model)
Error resolution priority — what models fix first
Percentage of each error type resolved between iteration 1 and final iteration (best run per model)
Score consistency — reliability across runs
Min, median, and max final score per model. Narrow range = reliable. Only models with 2+ runs shown.
First-iteration score — inherent WCAG knowledge
Average score on iteration 1 (before any feedback). Higher = better baseline knowledge.
Convergence speed — iterations to reach 90% of final
Average number of iterations to reach 90% of the model's final score (best run per model)