Output Loop Bench · Artifacts
Web page: AI models compared
One self-contained HTML page from a standard brief. 31 models tested on the same 2 tasks.
Electrician website with quote form
Trades are our first outreach. A strict palette, set sections and a working form show which models follow real art direction instead of falling back on a template.
Electrician website with quote form: all 30 results
| Model | Maker | Time | Tokens |
|---|---|---|---|
| muse-spark-1.3-medium | Meta | 18.4s | 25,829 |
| nemotron-3-nano-4b | NVIDIA | 33.3s | 3,783 |
| composer-2.5 | Cursor | 34.4s | 26,228 |
| kimi-k3-max | Moonshot (Kimi) | 55.6s | 29,393 |
| gemma-4-26b-a4b-qat | 57.2s | 7,219 |
Local business landing page
Everyday small-business work: a one-page site for a local shop. The brief has exact facts, so you can see whether a model sticks to them, and the page shows design, layout and phone support at a glance.
Local business landing page: all 25 results
| Model | Maker | Time | Tokens |
|---|---|---|---|
| gpt-6-astra ★ Output Loop's pick | OpenAI | 87.6s | 17,836 |
| muse-spark-1.3-medium | Meta | 11.0s | 22,059 |
| nemotron-3-nano-4b | NVIDIA | 12.5s | 1,463 |
| grok-4.7-build-fast | xAI | 13.0s | 18,915 |
| composer-2.5 | Cursor | 14.7s | 21,893 |