Last March, I got my credit card bill and noticed something strange: $20 to OpenAI. Again.
I'd been paying for ChatGPT Plus for eight months straight. Some months I used it constantly. Other months—like when I was on vacation in July, or slammed with client work in September—I barely touched it.
But OpenAI didn't care. $20. Every. Single. Month.
So I did what any reasonable person would do: I spent six months obsessively tracking every single ChatGPT query I made, calculated the actual cost, and compared it to what I was paying.
The results surprised me.
The Subscription Trap Nobody Talks About
Here's the thing about subscriptions: they're designed to extract maximum revenue with minimum friction.
You sign up in a moment of excitement. You use it heavily for two weeks. Then life happens. You use it less. Way less. But you keep paying because:
- Canceling feels like giving up
- "What if I need it next week?"
- It's only $20, right?
Wrong.
Over six months, I paid OpenAI $120 for ChatGPT Plus. Sounds reasonable until you look at my actual usage.
My Real Usage (The Uncomfortable Truth)
Here's what six months of data revealed:
March: 47 queries → Would've cost ~$3.50 with API pricing April: 134 queries → Would've cost ~$8.20 May: 89 queries → Would've cost ~$5.80 June: 12 queries (vacation) → Would've cost ~$0.90 July: 203 queries (new project) → Would've cost ~$14.50 August: 56 queries → Would've cost ~$4.10
Total paid: $120 What I actually used: ~$37 worth of compute Money wasted: $83 (69% of my spending)
I basically donated $83 to OpenAI for access I didn't use.
The Psychology of "Unlimited"
ChatGPT Plus markets itself as "unlimited" access. But that's a lie by omission.
You're not limited by queries—you're limited by time. You still have 24 hours in a day. You still have work, family, other responsibilities. The ceiling isn't technological, it's biological.
And OpenAI knows this.
The average ChatGPT Plus subscriber uses about 30-50 queries per day during their active periods. But most people aren't active every single day. They have:
- Weekends where they don't work
- Vacations where they unplug
- Busy periods where they use other tools
- Slow months where they just don't need AI help
You're paying for 30 days of access but realistically using maybe 18-20 days. That's 10 days per month of paying for nothing.
When the Math Actually Works
Look, I'm not here to trash subscriptions entirely. There are people for whom ChatGPT Plus makes financial sense.
If you're sending 80+ queries per day, every single day, you're probably getting your money's worth. At API pricing, that's about $0.60-0.80 per day, or $18-24 per month. ChatGPT Plus's $20 flat rate becomes a deal.
But be honest: Are you really that user?
Most of us aren't. Most of us are:
- Using it intensely during project crunch times
- Barely touching it during slow periods
- Supplementing with Claude, Perplexity, or other tools
- Taking breaks when we're traveling or on vacation
For us, subscriptions are a tax on unpredictable usage.
The Real Cost of Tool Sprawl
Here's where it gets worse. ChatGPT Plus isn't your only subscription, is it?
If you're anything like me, you've accumulated:
- ChatGPT Plus: $20/month
- Claude Pro: $20/month
- Perplexity Pro: $20/month
- Midjourney: $30/month
- Maybe GitHub Copilot: $10/month
That's $100/month. $1,200/year. On AI tools alone.
And the dirty secret? You're probably only actively using one or two at any given time. The others sit idle, auto-renewing, quietly draining your account.
I realized I was paying for optionality—the option to use these tools if I wanted to—rather than actual value delivered.
What Changed My Mind
In November, I ran an experiment. I canceled ChatGPT Plus and switched to using the API directly with a simple interface.
I prepaid $50 into my OpenAI API account. That was five months ago.
My current balance? $38.
I've spent $12 in five months for the exact same model (GPT-4) that ChatGPT Plus was charging me $100 for over that period.
Same quality. Same speed. 88% cost reduction.
The only difference? I pay for what I use. Nothing more.
The Uncomfortable Question
Why are we okay with this?
We've collectively accepted that software should be rented, not bought. That access should cost the same whether you use it once or a thousand times. That "simplicity" of billing justifies paying for waste.
But it doesn't have to be this way.
Usage-based pricing has existed in other industries forever. You don't pay a flat rate for electricity whether you use 10 kWh or 1000 kWh. Your water bill isn't $50/month regardless of consumption.
So why should software be different?
What Actually Makes Sense
After six months of data, here's what I learned:
Subscriptions make sense when:
- Your usage is genuinely high and consistent (50+ queries/day, every single day)
- You value billing simplicity over cost optimization
- You're bad at budgeting variable costs
- The subscription includes unique features (like DALL-E access in ChatGPT Plus)
Usage-based pricing makes sense when:
- Your usage varies month to month (most of us)
- You use multiple AI tools and want flexibility
- You take breaks from technology periodically
- You're cost-conscious and hate waste
- You want your spending to match your actual value received
For me—and I suspect for most people reading this—usage-based wins by a mile.
What I'm Doing Now
I've switched to a pay-per-use model across all my AI tools. No more subscriptions.
When I need GPT-4, I pay for GPT-4. When I don't, I pay nothing. Same with Claude, image generation, voice synthesis—everything.
My AI spending last month? $18. The month before? $31. The month before that? $7.
It varies with my actual needs. And that feels right.
I'm not locked in. I'm not paying for access I'm not using. I'm not subsidizing OpenAI's revenue goals at my expense.
I'm just paying for value received. Novel concept.
If You're Curious
I'm not saying cancel ChatGPT Plus tomorrow. But I am saying: track your usage.
Open up ChatGPT right now. Look at your actual activity. Count how many conversations you've had in the last 30 days.
If it's under 100, you're probably overpaying. If it's under 50, you're definitely overpaying.
Then ask yourself: Is the convenience of "unlimited" worth paying 2-3x what you actually use?
For me, it wasn't.
If you want to try a usage-based approach, platforms like ARES let you access multiple AI models (GPT-4, Claude, FLUX, and others) on a pay-per-use basis. Credits never expire, so you're never paying for time you don't use.
I'm not affiliated with them, but after my six-month subscription wasteland, tools like this make a hell of a lot more sense to me than renting access I don't fully use.
Got thoughts on subscription vs usage-based pricing? I'd genuinely love to hear them—especially if you've tracked your own usage and found different results.
