IWC-Bench: leaderboard
Interactive web-application generation judged from a software-testing perspective: an agent tester explores the generated app and scores aesthetics, usability, functionality, content and visual alignment; the total is standardized within the 17-model pool (z-score), main judge Claude-Opus-4.8. Each row names its harness (Claude Code or Codex) and reasoning effort in the paper. Higher is better. Source: arxiv.org/abs/2609.15387.
Metric: Standardized total score (z, judge Claude-Opus-4.8). Source: arxiv.org. Saturation forecast: Around July 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 1.33 | #16 |
| 2 | GPT-5.6 Sol | 1.08 | #22 |
| 3 | Qwen 3.8 Max | 0.97 | #34 |
| 4 | Kimi K3 | 0.97 | #35 |
| 5 | Hy4 preview | 0.91 | #45 |
| 6 | Claude Opus 4.8 | 0.76 | #39 |
| 7 | Grok 4.5 | 0.75 | #54 |
| 8 | GPT-5.5 | 0.47 | #33 |
| 9 | Claude Opus 4.7 | -0.26 | #51 |
| 10 | GLM-5.2 | -0.31 | #80 |
| 11 | DeepSeek V4 Flash | -0.35 | #155 |
| 12 | Hy3 | -0.59 | #100 |
| 13 | Qwen 3.7 Max | -0.91 | #85 |
| 14 | GLM-5.1 | -0.94 | #109 |
| 15 | Kimi K2.7 Code | -1.31 | #95 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=iwc-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.