MiniAppBench: leaderboard
Metric: Pass rate (%) over all 500 tasks, MiniAppBench's 500 interactive-HTML MiniApp tasks distilled from real user queries; the model writes a self-contained HTML/JS app in one generation under its official decoding settings, and the MiniAppEval browser agent (Gemini-3-Pro-Preview driving Playwright) scores Intention, Static and Dynamic fidelity from 0 to 1; a task passes when the minimum of the three scores is at least 0.8; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 45.46 | #105 |
| 2 | Claude Opus 4.5 | 41.14 | #79 |
| 3 | GPT-5.1 | 32 | #131 |
| 4 | Gemini 3 Pro (Preview) | 27.52 | #64 |
| 5 | Claude Sonnet 4.5 | 26.36 | #138 |
| 6 | GLM-4.7 | 18.31 | #185 |
| 7 | Gemini 3 Flash | 17.62 | #93 |
| 8 | MiniMax-M2.1 | 17.12 | #283 |
| 9 | Grok 4.1 Fast (Reasoning) | 13.77 | #208 (Grok 4.1 Fast) |
| 10 | MiMo-V2-Flash | 12.48 | #359 |
| 11 | GLM 4.5 Air | 7.09 | #377 |
| 12 | Kimi K2 | 6.19 | #236 |
| 13 | Qwen 3 235B A22B | 2.88 | #304 |
| 14 | Qwen 3 Coder 480B A35B Instruct | 1.83 | #302 |
| 15 | Qwen 3 32B | 0.66 | #424 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=miniappbench · How It Works · Data refreshed daily, snapshot 2026-10-11.