If you run a business in 2026, there's a good chance you're paying for AI. Maybe it's the $20-a-month ChatGPT subscription. Maybe it's bundled into your email or your design tools. Maybe you've just been clicking "upgrade" because you kept hitting some limit you didn't understand.
Here's the strange part: almost nobody who pays for AI can tell you what they're actually buying. What's a "token"? Why do "credits" run out? Why is there a $20 plan, a $100 plan, and a $200 plan — and what on earth justifies the expensive one?
You don't need to be technical to run a business well, but you should understand what you're paying for. So here's how AI actually works, in plain English — and which tier you probably need.
How AI Actually Works (The 60-Second Version)
Modern AI tools like ChatGPT, Claude, and Gemini are built on something called a large language model — an "LLM." Strip away the jargon and the idea is almost embarrassingly simple:
An LLM predicts the next word.
That's it. It was trained by reading a staggering amount of text — books, websites, articles, conversations — until it got extremely good at one task: given some words, guess what word comes next. Ask it "The capital of France is" and it predicts "Paris." String millions of those predictions together, one word at a time, and you get something that reads like a thoughtful human wrote it.
It isn't looking up answers in a database, and it isn't "thinking" the way you do. It's an extraordinarily sophisticated pattern-completion engine. That single fact explains almost everything else — why it's so fluent, why it occasionally makes things up with total confidence, and (importantly for your wallet) why it costs money to run.
What a "Token" Is — and Why You're Billed By It
AI doesn't actually read words. It reads tokens — small chunks of text. A token is roughly four characters, or about three-quarters of a word. "Hamburger" might be two tokens. "The" is one. A rough rule of thumb: 100 words is about 130 tokens.
Every single thing you type into an AI gets chopped into tokens, and every word it writes back is built token by token. And here's the key: the AI has to do real computation for every token, going in and coming out. That computation runs on expensive specialized chips in data centers that burn real electricity.
So when you hear "tokens" or "credits," just translate it in your head to: how much work am I asking the AI to do? A one-line question is cheap. Asking it to read a 40-page contract and summarize it is expensive — that's a lot of tokens going in. Asking it to write a 3,000-word report is expensive too — a lot of tokens coming out.
This is also why your AI sometimes "forgets" things from earlier in a long conversation, or slows down and cuts you off. You're bumping into how many tokens it can hold and handle at once.
Tokens vs. credits — what's the difference?
You'll see both words, and they mean slightly different things:
- Tokens are the raw unit, mostly used when you pay per-use (the "API" — what developers like me build apps on top of). You're billed for exactly what you use, often fractions of a cent at a time.
- Credits are a simplified, prepackaged version for everyday users. Image and video generators love this model: "you get 100 credits a month, each image costs 5." Same idea — a budget for how much computing you get — just dressed up in friendlier units.
Either way, the principle is identical: more sophisticated output = more computation = more tokens or credits burned.
The Tier Ladder: What $0, $20, $100, and $200 Actually Buy
By 2026 the major AI tools have settled into a remarkably similar pricing ladder. Prices shift constantly, so treat these as the shape of things rather than gospel — but as of mid-2026 it looks like this:
Free ($0)
You get real access to a capable model, but on a leash: daily message caps, slower responses when servers are busy, smaller memory, and limited or no access to the fanciest features. Free is genuinely useful for occasional questions. You'll feel the walls the moment you start leaning on it for real work.
The $20 Standard Tier
This is the sweet spot almost everyone lands on — ChatGPT Plus, Claude Pro, Gemini's Pro plan, and Perplexity Pro all hover right around $20/month. You get the flagship models, much higher usage limits, voice mode, image generation, file uploads, and the "reasoning" modes. For 95% of small business owners, this is the right tier and the conversation should honestly end here.
The ~$100 Power Tier
Plans like ChatGPT Pro and Claude's Max tier sit around $100/month and mostly buy you more: roughly 5x the usage limits, priority access, and first dibs on new features. You're not getting a smarter AI so much as a lot more of it. This tier is for people who use AI for hours every single day — developers, heavy researchers, agencies.
The ~$200 "Pro" Tier
At the top (~$200/month) you get the highest limits available — think 20x the standard plan — plus the most powerful reasoning models that literally "think longer" before answering, and early access to experimental tools. Unless AI is your full-time job, you almost certainly don't need this. It exists for power users and companies, not for the average owner deciding whether to upgrade.
Why "More Credits" Really Does Mean "More Sophisticated"
Here's the part that surprises people: spending more doesn't just remove limits — it unlocks genuinely different kinds of capability. Think of it as a ladder of sophistication, where each rung costs more computation:
- Text — the baseline. Type a question, get a written answer. Cheap, fast, available even on free tiers.
- Voice / actual talking — you can now have a real spoken conversation with AI, out loud, in real time, and it talks back in a natural voice. This costs more because the AI is converting your speech to text, thinking, and generating speech, all live. (I wrote a whole piece on why voice changes everything.)
- Images — describe a picture and it generates one. Each image is a big burst of computation, which is exactly why image tools almost always run on credits.
- Video — the newest and most expensive frontier. Tools can now generate short video clips from a text description. This eats credits fast, because video is essentially hundreds of images stitched together.
- Reasoning & agents — the top of the ladder. "Reasoning" models deliberately spend more computation working through a hard problem step by step before answering. "Agents" go further and actually do multi-step tasks for you — browse the web, fill out forms, write and run code. More thinking equals more tokens equals more cost, which is why these live on the higher tiers.
So the instinct is exactly right: the more you spend, the more sophisticated the things you can do — from a simple text reply, up to an AI that talks with you, makes pictures and video, and carries out tasks on its own.
The Tools People Are Actually Using in 2026
The space moves fast, but a handful of names have risen to the top. Here's the honest lay of the land:
- ChatGPT (OpenAI) — the household name and still the default for most people. Best all-rounder, strong voice and image generation, and a huge ecosystem of add-ons.
- Claude (Anthropic) — a favorite for writing, analysis, and anything involving long documents or code. (Full disclosure: it's what I lean on most for development work.)
- Gemini (Google) — deeply tied into Google's world: Search, Gmail, Docs, Android. Strong at images and video, and Google offers a cheaper entry plan around $8/month. If you live in Google Workspace, it's the natural fit.
- Perplexity — less a chatbot, more an AI-powered research engine. It answers questions with live web sources cited, which makes it great for fact-finding. (It's also part of why showing up in AI search now matters for your business.)
- Grok (xAI) — built into X (formerly Twitter), known for a looser personality and real-time access to what's happening on the platform.
- Microsoft Copilot — Microsoft's AI woven into Windows and Office, so it shows up right where a lot of businesses already work.
For images and video specifically, tools like Midjourney, OpenAI's image models, and Google's video models do the heavy lifting — almost all of them on credit systems.
Which Tier Does a Small Business Owner Actually Need?
Cutting through all of it, here's my honest advice:
- Just curious or occasional use? Start free. Genuinely. Learn what it's good at before you spend a cent.
- Using it for real work most weeks — drafting emails, writing content, brainstorming, summarizing documents? Pay the $20 for one tool and commit to learning it well. One tool you know deeply beats five you barely poke at.
- Living in AI for hours a day, or running a team on it? The $100 tier starts to make sense. Very few owners are actually here.
- The $200 tier? If you have to ask, you don't need it.
And choose based on where you already work. In Google all day? Gemini. Writing and documents? Claude. Want the safe, popular all-rounder? ChatGPT. You don't need all of them — you need the one that fits your day.
My Take
The pricing feels confusing on purpose, but the underlying idea is simple and worth holding onto: you're paying for computation. Tokens, credits, tiers — they're all just different ways of metering how much work you're asking a very expensive machine to do on your behalf. A quick question is pennies. A talking assistant that generates images, writes reports, and runs tasks for you is a lot more — and now you know exactly why.
For nearly every small business owner reading this, the move is the same: pick one tool, pay the $20, and actually learn it. That single subscription, used well, will do more for your business than the $200 plan you were nervous about — and far more than the free one you keep hitting the ceiling on.
Not sure how AI actually fits into your business — which tool, which tier, or where it would genuinely save you time? Send me a message and I'll give you a straight, no-hype answer for your specific situation.