Just a quick question about tokens vs subscription

Hi all, I’m a complete noob when it comes to using chatgpt, so sorry for asking the obvious … but am I right in thinking if I sign up & pay the $20 a month subscription fee, it just enables me to use it’s additional features & plugins, but I still have to pay for however many characters / words / tokens I use?

Tokens are a concept only applicable to the API.

If you’re paying $20/month and using ChatGPT, then that’s all you need to think about.

If you want to use the API and build your own applications with the OpenAI models on the backend, then you need to think about tokens and pay for every token you use.

So in terms of using chatgpt to write stuff for me, if I’m paying the €20 subscription fee, there’s no actual limit in regards to the number of characters / words per day? … Basically, I’ll just be using it to write (at max) 4 articles or blog posts a day, with a maximum of say, 2,000 words per post & I’m trying to work out the cheapest option, tokens or monthly subscription?

The only limits you’ll run into will be the results of being limited to 25-messages every 3-hours if you’re using GPT-4.

If you’re concerned about it, I would use 3.5 as much as possible to create outlines and rough-drafts of the posts first, then pass the draft to GPT-4 to re-write and polish the article.

In terms of the cheapest path, that’s fairly straightforward to calculate if you’re willing to make some assumptions.

It’s only slightly complicated by the fact that input tokens are cheaper than output tokens. Aggregated you can probably expect the overall token cost to be about $0.04 / 1k tokens for GPT-4. $20 would buy you 500,000 GPT-4 tokens.

A 2,000-word article is roughly 2,700 tokens.

With the back and forth, I imagine total token cost for a 2,000-word article would be in the vicinity of 25,000 tokens. So you could likely generate 20 articles for $20.

You’d need to develop a very efficient system of prompts to generate 120-articles of high-quality each month using GPT-4 through the API and limiting your cost to $20/month.

GPT-3.5 costs less than 1/20 what GPT-4 does, so you could almost certainly make 120-articles each month for ~$10 or so if you were content using GPT-3.5.

In the end, I suppose it comes down to time and number of prompts.

If you do most of the scut-work in GPT-3.5 and just do a final pass in GPT-4 you wouldn’t need to worry about the message-cap as much in ChatGPT, whereas if you tried to do it all in GPT-4 you’d almost certainly burn through the 25 messages and need to wait for your counter to reset.

While in the API, doing just a final pass with about 2,700 tokens in and out for each article comes to around 22,000 tokens each day at an average of $0.045 / 1k tokens which amounts to ~$30 / month.

What the API gives you is more control and fewer limits which can ultimately lead to better results.

tl;dr: ChatGPT will almost certainly be the more economical answer unless you exclusively use GPT-3.5.

Thanks for all the info, that’s just what I needed.