Recent advances in large language models have improved the capabilities of coding agents, yet systematic evaluation of complex, end-to-end website development remains limited. To address this gap, we introduce Vision2Web, a hierarchical benchmark for visual website development, spanning from static UI-to-code generation, interactive multi-page frontend reproduction, to long-horizon full-stack website development. The benchmark is constructed from real-world websites and comprises a total of 193 tasks across 16 categories, with 918 prototype images and 1,255 test cases. To support flexible, thorough and reliable evaluation, we propose workflow-based agent verification paradigm based on two complementary components: a GUI agent verifier and a VLM-based judge. We evaluate multiple visual language models instantiated under different coding-agent frameworks, revealing substantial performance gaps at all task levels, with state-of-the-art models still struggling on full-stack development.
Figures & tables
Figure 1: Overview of Vision2Web, a hierarchical benchmark for visual website development. Tasks span three levels—static webpages, interactive frontends, and full-stack websites—requiring agents to integrate visual prototypes with textual specifications. Evaluation is performed via a workflow-based agent verification paradigm, measuring functional correctness and visual fidelity.
Benchmark
# Tasks
# Test Cases
# Prototypes
# Multimodal
Task Types
SWE-Bench Multimodal
617
−
−
✓
Issue fixing
Design2Code
484
−
484
✓
Single page UI-to-Code
WebGenBench
101
647
−
✗
Website generation
Vision2Web
193
1255
918
✓
Full-stack website development
Table 1: Comparison of Existing Benchmarks for Software Engineering and Web Development Tasks.
Figure 2: Task distribution of Vision2Web across four major categories and 16 subcategories.
Level
Webpage
Frontend
Website
Avg Prototype Images
3±0
5.9±0.5
8.5±2.3
Avg Test Cases
−
7.5±4.5
28.2±12.0
Avg Text Tokens( 103 )
−
1.0±0.6
4.3±1.1
Table 2: Average prototype images, test cases, and input text tokens for Vision2Web across three task levels.
Figure 3: Distribution of test cases across website-level tasks in Vision2Web.
Coding Agent
Static Webpage
Interactive Frontend
Full-Stack Website
Desktop
Tablet
Mobile
Avg
DSR
VS
FS
Avg
DSR
VS
FS
Avg
DSR
Claude Code
Claude-Opus-4.5
54.2
50.4
46.8
50.5
94%
46.1
63.6
54.9
98.5%
34.3
53.1
43.7
92.6%
Claude-Sonnet-4.5
44.7
38.6
36.7
40.0
89%
24.9
48.7
36.8
93.9%
14.5
26.8
20.7
66.7%
GPT-5
45.3
44.2
39.9
43.1
97%
23.3
64.7
44.0
98.5%
9.6
23.4
16.5
85.2%
Gemini-3-Pro-Preview
59.2
51.5
46.2
52.3
92%
13.3
14.6
14.0
65.2%
5.7
12.9
9.3
63.0%
Table 3: End-to-end performance of multimodal coding agents on Vision2Web across three task levels, reporting device-wise static scores, averaged functional scores (FS) and visual scores (VS) for interactive and full-stack tasks, with Deployment Success Rate (DSR) provided for reference rather than an official metric. Unless otherwise noted, all metrics are reported on a 0–100 scale.
Figure 4: Distribution of Visual Scores (VS) across prototype heights for representative models under the OpenHands framework.
Website Category
Opus-4.5
Sonnet-4.5
GPT-5
Content
37.1 / 61.2
9.3 / 16.1
20.7 / 53.5
Transaction
43.2 / 64.9
10.8 / 14.3
13.4 / 50.6
SaaS Platform
22.9 / 39.9
21.7 / 42.8
16.7 / 40.5
Public Service
56.9 / 60.0
41.2 / 52.0
27.4 / 56.0
Table 4: Performance (Visual Score / Functional Score) of selected coding agents across different website categories under the OpenHands framework in Vision2Web. Opus-4.5 and Sonnet-4.5 refer to Claude-Opus-4.5 and Claude-Sonnet-4.5 respectively.
Test Case Category
Opus-4.5
Sonnet-4.5
GPT-5
Navigation & Routing
66.3
25.9
53.9
State Management
43.2
16.1
41.5
Form Interaction
49.2
23.7
56.8
UI Feedback
56.4
23.1
30.8
Authentication & Authorization
61.8
25.5
60.0
CRUD Operations
43.6
20.5
35.9
Table 5: Performance of selected coding agents under the OpenHands framework on Vision2Web, reporting category-wise functional scores (average test-case pass rates).
Tested Model
GLM-4.6V FS
GPT-5-Mini FS
Claude-Sonnet-4.6
48.6
46.6
GPT-5.4
28.6
29.3
Gemini-3.1-Pro-Preview
17.0
19.1
Overall
31.36
31.67
Table 6: Sensitivity of functional verification to GUI verifier backbones on Level-3 website tasks. FS denotes Functional Score.
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
Macro Category
Subcategory
Description
Examples
Content
News
Platforms for publishing timely news content
CNN, BBC, NYTimes
Community
Platforms for content and social interaction
Reddit, Zhihu, StackOverflow
Multimedia
Platforms for consuming rich media content
YouTube, Spotify, Vimeo
Knowledge
Platforms for knowledge delivery
Coursera, Khan Academy, edX
Publishing
Platforms for content creation and presentation
Medium, WordPress, Substack
Transaction
Retail
Platforms for consumer product transactions
Amazon, Taobao, JD
Appendix
Table 7: Website categories in Vision2Web, subcategory definitions, and representative examples.
Figure 12
Macro Category
Subcategory
Static Webpage
Interactive Frontend
Full-Stack Website
Content
News
8
7
2
Community
8
6
2
Multimedia
6
5
2
Knowledge
10
7
2
Publishing
8
5
2
Transaction
Retail
10
5
2
Appendix
Table 8: Distribution of tasks across website categories and development levels in Vision2Web.
Web agent benchmarks test how reliably AI agents browse, navigate, and act on websites, including under long-horizon, ambiguous, and adversarial conditions.