This is my second post on this topic. The first post is here ChatGPT Plus, Claude Pro, vs API usage

Per my calculations, the $20 subscriptions of ChatGPT and Claude provide incredible value for money. Let's quantify it.

Claude Pro ($20 per month)

The Claude Usage Extension, that has been quite accurate in my usage, tells us that the usage limit for Claude Pro resets every 5 hours.

Every 5 hours, you get:

• 1.6 million tokens for Sonnet
• 500K tokens for Opus
• 3 million tokens for Haiku

This is a total across input and output, where output is weighed at 10x compared to input. That means 1K output tokens count as 10K tokens used, while 1K input tokens count as 1K tokens used.

Since a 3:1 input to output blend is typical in calculating blended costs for LLM APIs, we'll stick with that ratio.

Based on the numbers, this is how much we can use for each model per 5 hours.

So every 5 hours, if you were maxing out your account, you would be able to use $4.61 + $2.96 + $1.47 = $9.04 worth of AI usage (if you used the same amount via API). I am ignoring prompt caching for simplicity. There are technically 4.8 5 hour periods in a day, so if you maxed out each of those 5 hour periods, you'd spend ~$43.4 worth. For a typical 30 day month, that's ~$1300.

$1300. That's the theoretical MAX value that you could extract out of that $20 plan vs. using the API directly.

Even if you aren't maxing it out, even just 10% of that is still $130, which is 6.5x of what you paid.

ChatGPT Plus ($20 per month)

ChatGPT is more complicated due to the additional services they provide like Advanced Voice Mode, Deep Research, etc.

Looking at just the models though. Per OpenAI

As of Feb 1, 2025, Plus users will be able to send up to 80 messages every 3 hours on GPT-4o and up to 40 messages every 3 hours on GPT-4. We may reduce the limit during peak hours to keep GPT-4 and GPT-4o accessible to the widest number of people.

Source: https://help.openai.com/en/articles/7864572-what-is-the-chatgpt-model-selector

So 80 messages to GPT-4o and 40 messages to GPT-4 every 3 hours. Both of these models have up to 32K token context window on ChatGPT Plus.

You also get access to their "o" series reasoning models.

For o3-mini, you get 150 messages per day.

o3‑mini will replace OpenAI o1‑mini in the model picker, offering higher rate limits and lower latency, making it a compelling choice for coding, STEM, and logical problem-solving tasks. As part of this upgrade, we’re tripling the rate limit for Plus and Team users from 50 messages per day with o1‑mini to 150 messages per day with o3‑mini.

Source: https://openai.com/index/openai-o3-mini/

And for o1, you get 50 messages per week.

With a ChatGPT Plus, Team or Enterprise account, you have access to 50 messages a week with OpenAI o1 and 50 messages a day with OpenAI o1-mini.

Source: https://help.openai.com/en/articles/9824962-openai-o1-o1-mini-and-o3-mini-usage-limits-on-chatgpt-and-the-api

Let's look at GPT-4o and GPT-4, we'll skip GPT-4o mini for now.

Assuming the same 3:1 input:output token blend, and assuming we absolutely max out the context window on each message, that'll be 24K input tokens and 8K output tokens per message.

GPT-4o in the API costs $2.5/$10 per mn input / output tokens, whereas GPT-4 costs $30/$60 for the same.

So every 3 hours, if you were maxing out both the GPT models, you'd be using $59.2 worth of AI vs. if you used it via the API.

If you absolutely maxed these out, you'd use 240x the above as there are 240 3 hour periods in the average month.

The math for that is as follows:

Let's look at the cost for o1 and o3-mini, for just one message below.

I want to clarify, OpenAI says the context window for o1 and o3-mini on the Plus plan is 32K tokens but there's some ambiguity there in my experience, but let's just stick with 32K tokens for now.

If you max out your quota for the reasoning models, you'd use 4500 messages a month on o3 mini and 21o messages a month on o1.

Advanced Voice Mode is super expensive too on the API, more than GPT-4o on its own, and Deep Research, but it won't change the bottom line here.

You are getting $2688 + $11520 + $176 + $277 = $14,661 worth of usage for your $20 every month.

If you only used 1% of that, that's still $146 worth of usage, or 7.3x of what you paid!

Conclusion

The $20 per month AI plans are bonkers value for money. People are quick to suggest using the API to "save money", but, would you, really? If you were using it heavily, absolutely not. Not to mention the hidden cost that seeing your API usage bill go up and doing mental arithmetic with every single message you send. Will that hinder your usage? Would you use it more, and hence, save more time and get more value out of it, if you knew it was a fixed cost? In my experience that's absolutely true.