ChatGPT Plus, Claude Pro, vs API usage
ChatGPT Plus and Claude Professional both cost $20 per month.
These are the consumer interfaces for these AI tools, but you can also access these AI models via their APIs. You have many frontend options, like TypingMind, LibreChat, and OpenWebUI, among others.
So what does $20 of API credits get you?
Assumptions
Let's get some assumptions out of the way about what we expect each chat to use in terms of tokens.
All AI models charge per token, and the pricing for output tokens is always higher compared to input tokens. You can consider tokens ~= words, 1000 words are 750 tokens.
Since the input and output tokens are charged differently, this can complicate our calculations, but generally, a 3:1 blend of 3 input tokens for each output token is widely used, so we'll stick with this.
How this works is, for every 1Mn tokens, 750K of the tokens would be input tokens, and 250K would be output tokens.
Based on my own usage, I can see an average chat using ~15K input tokens. Following the 3:1 blend, we'll take 5K as the average output token usage per chat (actual numbers will vary).
ChatGPT
gpt-4o
Let's start by looking at gpt-4o
(the latest variant, gpt-4o-2024-08-06
)
The model stated above costs $2.5 per Mn input tokens, and $10 Per Mn output tokens.
(750,000/1,000,000) x 2.5 = 0.75 x 2.5 = 1.875 USD
(250,000/1,000,000) x 10 = 0.25 x 10 = 2.5USD
Blended cost = $1.875 + $2.5 = $4.375
Now on a very basic level, this gives us ~4.5Mn tokens worth of usage of gpt-4o-2024-08-06
).
Chat cost
With our assumption of 15K input and 5K output tokens per chat, each chat would cost us
(15,000/1,000,000) x 2.5 = $0.0375 for input tokens
(5,000/1,000,000) x 10 = $0.05 for output tokens
Total cost per chat = $0.0875
For $20, you can have ~230 chats
This is almost 8 long chats per day
gpt-4o-mini
gpt-4o-mini
costs $0.15 per Mn input tokens, and $0.6 per Mn output tokens.
For every 1Mn tokens, 750K of the tokens would be input tokens, and 250K would be output tokens.
(750,000/1,000,000) x 0.15 = 0.75 x 0.15 = 0.1125 USD
(250,000/1,000,000) x 0.6 = 0.25 x 0.6 = 0.15 USD
Blended cost = $0.1125 + $0.15 = $0.2625
This gives us ~76Mn tokens worth of usage of gpt-4o-mini
Chat cost
With our assumption of 15K input and 5K output tokens per chat, each chat would cost us
(15,000/1,000,000) x 0.15 = $0.00225 for input tokens
(5,000/1,000,000) x 0.6 = $0.003 for output tokens
Total cost per chat = $0.00525
For $20, you can have ~3800 chats
This is almost 127 long chats per day
Claude
Let's examine the pricing for Claude models, starting with Claude 3.5 Sonnet.
Claude 3.5 Sonnet
Claude 3.5 Sonnet costs $3 per Mn input tokens and $15 per Mn output tokens.
Using our 3:1 blend of input to output tokens:
(750,000/1,000,000) x 3 = 0.75 x 3 = 2.25 USD for input
(250,000/1,000,000) x 15 = 0.25 x 15 = 3.75 USD for output
Blended cost = $2.25 + $3.75 = $6
This gives us ~3.33Mn tokens worth of usage of Claude 3.5 Sonnet.
Chat cost
With our assumption of 15K input and 5K output tokens per chat, each chat would cost us:
(15,000/1,000,000) x 3 = $0.045 for input tokens
(5,000/1,000,000) x 15 = $0.075 for output tokens
Total cost per chat = $0.12
For $20, you can have ~167 chats
This is about 5-6 long chats per day
Claude 3 Haiku
Claude 3 Haiku costs $0.25 per Mn input tokens and $1.25 per Mn output tokens.
Using our 3:1 blend: (750,000/1,000,000) x 0.25 = 0.75 x 0.25 = 0.1875 USD for input (250,000/1,000,000) x 1.25 = 0.25 x 1.25 = 0.3125 USD for output
Blended cost = $0.1875 + $0.3125 = $0.5
This gives us ~40Mn tokens worth of usage of Claude 3 Haiku.
Chat cost
With our 15K input and 5K output tokens per chat: (15,000/1,000,000) x 0.25 = $0.00375 for input tokens (5,000/1,000,000) x 1.25 = $0.00625 for output tokens
Total cost per chat = $0.01
For $20, you can have ~2000 chats
This is about 67 long chats per day
Conclusion
- If you are a light user, you will definitely save money by using the API, and $20 could last you 3+ months easily, bringing your monthly cost around $6 (assuming you use ~2 long chats per day with Claude 3.5 Sonnet, other models will be cheaper stilll)
- If you are a heavy user, keep in mind that whatever OpenAI and Anthropic are charging for API access to you, they're not doing it at a loss, which means that technically, you could end up spending more than $20 per month for the same usage as the official apps. But you can get more advanced features that are available in third-party clients like TypingMind.