LongWebBench - Functional Task Success (Single-Image): leaderboard
Functional task success on LongWebBench (2026) for webpages generated from one long screenshot of each page: the percent of 507 goal-oriented interaction tasks over 129 long webpages that a Gemini-3-Pro actor agent completes on the generated page, as checked by a Gemini-3-Pro critic. Higher is better.
Metric: Task success rate (%) on LongWebBench's 507 goal-oriented interaction tasks over 129 long webpages generated from one long screenshot of each page: a Gemini-3-Pro actor agent executes each task's steps in a browser on the generated page and a Gemini-3-Pro critic checks the DOM and rendered state; a task succeeds only if every step is executable and the expected outcome is reached; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 (Thinking) | 59.4 | #250 (Claude Sonnet 4.5) |
| 2 | Gemini 3 Pro (Preview) | 58.38 | #113 |
| 3 | GLM-4.6V | 28.39 | #701 |
| 4 | Qwen 3 VL 235B A22B Instruct | 23.65 | #522 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=longwebbench-functional-task-success-single-image · How It Works · Data refreshed daily, snapshot 2026-11-13.