Build a website

“I want a website and I do not write code”

What one run costs

A typical run of this task sends about 45K tokens and gets back 6K, over 40 calls — 2.0M tokens in total. Those are the numbers below, against the published prices.

Ordered by what a run costs. A model that does not publish a figure this task needs is listed last and says so, rather than being left out silently.
Model	Vendor	Per run	Can do what this needs?	Price evidence
GPT-5 Nano	OpenAI	$0.1860	not stated	Vendor page
Gemini 2.5 Flash Lite	Google	$0.2760	not stated	Vendor page
GPT-4.1 nano	OpenAI	$0.2760	not stated	Corroborated
Claude Haiku 5.5	Anthropic	$0.3000	confirmed	Vendor page
GPT-6 Luna	OpenAI	$0.3000	not stated	Vendor page
GPT-4o mini	OpenAI	$0.4140	not stated	Corroborated
Jamba Mini	AI21	$0.4560	not stated	Vendor page
GPT-5.6 Luna	OpenAI	$0.6480	not stated	Vendor page

Forty or so turns, each one re-sending the files it has already seen. That shape is why the cheapest-looking model is often not the cheapest session: what you pay for is mostly context being sent over and over, not the code coming back.

The practical order is to start with a mid-priced model, let it scaffold the thing, and switch up only for the parts it gets visibly wrong. Switching down is harder than switching up, because by then the context is long.

What goes wrong is rarely code that does not run — that is obvious in seconds. It is code that runs and is quietly wrong: a form that does not validate, a page that leaks something, an edge case nobody tried. Decide up front which parts you are still obliged to read.

What we would pick

This section is our judgement, not a figure read off a page. Everything above is arithmetic on published prices; this is an opinion, and it is labelled as one.

ShareOpen LinkedIn