Cookie-Bench - HTML: leaderboard

Cookie-Bench (2026) tests LLMs on front-end web development over 1,000 queries in 11 domains, with an autonomous evaluator that operates each generated page and scores the recorded session for functionality and aesthetics against human-calibrated rubrics, on a 0 to 100 scale. This board is the overall score when the model writes one self-contained HTML file directly from the query, without a scaffold or tools; higher is better.

Metric: Cookie total score (0-100) from the autonomous Cookie evaluator, which drives the rendered page and judges the continuous execution video against human-calibrated functionality and aesthetics rubrics, overall over the 1,000 queries, HTML mode: the model receives only the query and returns one self-contained HTML file, with no scaffold or tools; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.784.2#85
2Claude Opus 4.678.9#113
3Kimi K2.678.9#192
4GPT-5.478.6#135
5GLM-5.176.1#198
6DeepSeek V4 Pro74.6#196
7Qwen 3.6 Plus74.4#238
8MiMo-V2-Pro73.9#353
9Gemini 3.1 Pro (Preview)73.5#99
10DeepSeek V4 Flash72.6#267
11Qwen 3.6 27B70.7#351
12Gemini 3 Flash66.1#175
13Qwen 3.6 35B A3B63.1#481

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=cookie-bench-html · How It Works · Data refreshed daily, snapshot 2026-11-21.