This page in plain text
Which Claude model should you use?
Claude comes in three sizes: Opus, the largest; Sonnet, the middle one; and Haiku, the small fast one. The larger the model, the more each answer costs. Before we chose a model for the chat on this site, we asked all three the same ten questions.
They were questions an ordinary reader might ask, sent through our own chat on a short page about writing prompts, with the same instructions each time.
The numbers
| Model | Cost per answer | Cost per 200 answers | Typical wait | Average length |
|---|---|---|---|---|
| Opus 5.5 | $0.0070 | $1.40 | 4.0 s | 127 words |
| Sonnet 5.5 | $0.0034 | $0.69 | 3.5 s | 119 words |
| Haiku 5.5 | $0.00014 | $0.03 | 1.9 s | 118 words |
One Opus answer cost the same as fifty from Haiku, and Haiku answered in about half the time. Sonnet cost about half of Opus and was a little faster.
What we saw in the answers
- All three gave an answer we would be content to show a reader, on every one of the ten questions.
- All three told people to check. Asked about customer data, each said to ask the company first and strip out names.
- Opus and Sonnet more often said when the page did not cover a question before answering from general knowledge. Haiku more often just answered.
- Opus added the most detail, which was sometimes the useful part and sometimes just longer. Haiku’s replies were as long as the others and often the most direct.
What we chose
A guest’s questions on this site run on Haiku, and a member’s run on Sonnet. For short questions about a page, Haiku was good enough that paying fifty times more made no sense, and Sonnet gives members the more careful answer for about half the price of Opus.
What to choose yourself
- Haiku for short jobs done many times: answering questions about a page, sorting, tagging, quick summaries.
- Sonnet as the default for real work: writing, analysis, code.
- Opus for hard, long jobs with many steps, where one better answer is worth more than the price.
What this test does not show
Our questions were short and easy. Nothing here asked a model to plan a long task, work through a large document or write a program, which is where the larger models earn their price. Each question was run once, as a single message, on a page of about 150 words. Prices and models change, so treat the figures as a snapshot of November 2026.
Ask Another AI Guy
I'm just Another AI Guy.
You're chatting with our fast model. Create a free account for the more thorough one, and more messages.