I-WebGenBench: leaderboard
I-WebGenBench evaluates whether a model can generate an interactive scientific web application from a paper-derived specification across five domains (biology, chemistry, computer science, economics, human-computer interaction), with 201 tasks. This board reports Overall score (%): mean of a 100-point rubric (visual quality, interaction fidelity, topic alignment, clarity, rule and stability) averaged over 201 paper-derived interactive web-app tasks in five domains, built in a sandbox and DOM-probed; higher is better.
Metric: Overall score (%): mean of a 100-point rubric (visual quality, interaction fidelity, topic alignment, clarity, rule and stability) averaged over 201 paper-derived interactive web-app tasks in five domains, built in a sandbox and DOM-probed; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 79.4 | #123 |
| 2 | Gemini 3.1 Pro (Preview) | 53.8 | #94 |
| 3 | Kimi K2.5 | 48.9 | #256 |
| 4 | Qwen 3 Max | 31.5 | #403 |
| 5 | MiniMax-M2.7 | 30.4 | #518 |
| 6 | DeepSeek V4 Flash | 20.7 | #258 |
| 7 | MiMo-V2.5 | 16.5 | #354 |
| 8 | DeepSeek V3.2 | 9 | #406 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=i-webgenbench · How It Works · Data refreshed daily, snapshot 2026-11-13.