WebIGBench (Direct Prompt) - Action Completion Rate: leaderboard
Partial-credit score of WebIGBench (Wu et al., 2026) webpage generation under the Direct prompt (screenshots and textual instructions): the share of each page's reference interaction steps that an LLM matcher aligns with a step the UI agent executed on the generated page, averaged over 103 real-site pages. Percent, higher is better; temperature 0.
Metric: Action completion rate (%; share of each page's reference actions matched to an executed action on the generated page, mean over 103 webpages, screenshots and textual instructions). Source: arxiv.org. Saturation forecast: Around March 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 75.4 | #262 |
| 2 | Qwen-VL-Max | 61.64 | #1226 |
| 3 | GPT-5 | 56.36 | #163 |
| 4 | Grok 4 | 48.65 | #306 |
| 5 | GLM-4.5V | 46.55 | #739 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=webigbench-direct-prompt-action-completion-rate · How It Works · Data refreshed daily, snapshot 2026-11-12.