AKA12 / AI website benchmark
One driver. One task.
Every model’s result.
Build a website about Kimi Antonelli. Same task, different AI models. Explore what each one made and compare the time, cost and text used.
All test runs
Select two runs below to compare their websites side by side.
Cost is in US dollars. Text is counted in small pieces called tokens. These numbers do not measure website quality. Select a preview to open the website ↗
| Website preview | Model / run | Thinking setting | Test date | Reported cost | Time taken | Text used (tokens) | Website & details |
|---|
No matching runs
Try a different search or clear the filters to see every run.
What the numbers mean
Read the original task ↗Cost and time
Cost is the amount reported by the app. For subscription plans, it can be a price estimate rather than an extra charge. $0 means zero was recorded. “Not recorded” means the value is missing. Time covers the full test run.
Text used
Tokens are small pieces of text read or produced by a model. The total includes input and output across the run. Reused input is already included in input; thinking is already included in output. Neither is added twice.
Thinking setting
This is the requested amount of thinking: from none to maximum, or the model’s default. Settings can work differently across models. A higher setting, longer run or higher cost does not guarantee a better website.
How these tests work
Each model receives the same task and reference files, with one attempt and no follow-up conversation to fix the result. The reference race data is frozen on 12 October 2026. Run details include the app version when recorded, because software changes can affect results.
Every recorded attempt is listed separately. When attempts share a saved website path, they link to the same currently saved page; earlier versions are not available here. Notes are shown as recorded, including failed actions. This page has no design-quality scores.