Did you compare the cost of using a GPT plus subscription and just use GPT by API?

As a software engineer, mostly time I use Copilot(Chat) in my daily work.
It is very convenient because the Copilot chat is integrated with the VSCode.
I have been a GPT Plus user for a couple of months, but recently I canceled my subscription
because I found my frequency of GPT Plus is reducing.
Now there are a lot of open-source GPT clients, I made a client from one of them and deployed it on my server.
I think it is enough for me. So I started to use GPT by the ChatGPT API, its advantage is that I’m only charged as I use it.
I don’t know if it is a smart choice, but I’d like to hear your opinions.
Did you guys compare the cost of using GPT Plus and ChatGPT API?
Any suggestions are welcome.

1 Like

I like plus as it gives me access to gpt4 and the tools instead of 3.5 without the tools, in the web interface. I find i’m spending more than $20 monthly on the api’s mainly because Assistants api is quite expensive token wise, but i find it very useful.

2 Likes

Clarification: The models that you access via the API are not called “ChatGPT”, so it cannot be called ChatGPT API.

OpenAI also made calling something “GPT” very confusing with the new agent gizmo that comes as a feature of ChatGPT Plus.

So, (being able to understand what you are asking anyway),

yes, paying directly for model data usage by API can be advantageous, and you also have the advantage of being able to develop specific applications and control the quality of language generation with sampling parameters.

Having the option of choosing which is better for you is great, if you have the moderate skill to use the API via programming and keep your own data and authentication secure.

4 Likes

Normally for Co-Pilot stuff, which I don’t use too often, I would use the API with the Playground interface (built-in) and probably spend less than $1 to $5 per month. Which is cheaper than ChatGPT. This is for generic stuff.

Since you are a software developer, you could also build a RAG system based on whatever code base you want to draw from, and generate a Co-Pilot drawn from code that is tailored to your situation.

The benefits of the API route are pay-as-you-go (often cheaper) and you get much more than 40 messages of GPT-4 per 3 hours (so less restrictions if you are in a coding frenzy and need to iterate faster)

1 Like

I also have my own app that’s able to call thru the API, and so I have never been forced to signup for “Plus”, to get the GPT4.0. If I have a simple question for which I know ChatGPT 3.5 can suffice I’ll use the free web app on OpenAI website, and not even spend my own money.

For me Copilot is amazingly useful. It’s somehow able to guess a massive amount of the things I’m about to type, and is darn near clairvoyant in doing so. I can’t imagine how anyone would not find Copilot useful, unless they simply don’t know how to use it effectively.

1 Like