Start with the free plan. Always.
Don't reach for your credit card on day one. Sign up, start using it, and pay attention to what happens. The free tier isn't just a demo - it's a diagnostic. It tells you who you are as a user by showing you exactly where the walls are.
Because the walls will come. And how fast you hit them tells you everything about what you actually need.
Here's what the pricing pages leave out. When you hit a usage limit on an AI platform, it doesn't always put up a red light. Especially in the early days, there was no warning at all. What happened was subtler and more frustrating: the AI started to unravel.
Responses would get shorter. Less coherent. More generic. It would lose the thread of what you'd been building together - the context, the voice, the momentum. You'd think the tool was broken. It wasn't broken. It was rationed.
That's the threshold. And once you recognize it, you never mistake it for a malfunction again.
The platforms have gotten better about this. Claude now tells you when you're approaching a limit. ChatGPT has clearer messaging too. But the experience of hitting the wall mid-project - when you're deep in a writing session and the quality suddenly drops - that's still the moment most people decide to upgrade. Not because a pricing page convinced them. Because the work demanded it.
This is the question that determines everything: what are you building?
If you're asking the occasional question, brainstorming an idea, or kicking the tires on what AI can do - the free tier is fine. More than fine. It's plenty.
But if you're writing a book, developing a course, running heavy research, building systems - the free tier will cost you more in frustration than the paid tier costs in money. That's not a sales pitch. That's math.
I spend hours a day inside these tools. I moved to the paid plan within the first few weeks - not because someone told me to, but because I was hitting walls every session. Eventually I went to the Pro tier, and even there I still maxed out chat threads and found the edges. Because I pushed it hard. The tool matched my ambition, and the ambition outgrew the tool.
If you know going in that you're building something real, budget for the tool that can hold it.
One pattern before the price tags, because it makes the whole market legible: the industry converged on a shared ladder. A free tier. A twenty-dollar professional seat. Then a hundred dollars for roughly five times the usage, and two hundred for roughly twenty times. The expensive tiers overwhelmingly sell headroom, not new abilities. Hold that shape and every menu below reads in seconds.
Anthropic doesn't publish exact message caps for Claude. The limits are dynamic - they shift based on how long your messages are, how complex your requests are, and how much load the system is carrying. That's part of the game. But here's what the tiers actually look like in practice.
You get the current default model, Sonnet 5, on web, iOS, Android, and desktop. You can upload documents, analyze images, and do basic web searches. But the daily message cap is tight, and during peak hours it gets tighter. Think of it as a generous test drive with a curfew.
This is where the real work begins, and the annual price makes it the only sub-twenty seat among the majors. Pro unlocks the full model lineup - Haiku for quick tasks, Sonnet for speed with quality, Opus as the everyday deep thinker, and Fable when the work gets heaviest - plus Projects (more on that later), workspace integrations, extended thinking, memory across conversations, and Claude Code for developers.
One nuance worth knowing before you push hard: usage runs on two clocks. A rolling five-hour session window, and a weekly ceiling above it. Burst hard for two days straight and you can spend the week while the session meter still shows room. Pace the heavy lifts.
For most people doing serious creative work, Pro is the sweet spot. If you're using Claude three or more hours a day, this is where you belong.
Max comes in two flavors. The $100 tier gives you five times the Pro usage. The $200 tier gives you twenty times Pro, which effectively removes the concept of limits for most individual users. Both include priority access during peak traffic.
You need Max when Pro's limits interrupt your workflow more than once a week. If you're running multi-hour sessions on complex projects - building a book, managing a business through AI, coding full applications - Max gives you the room to work without watching the clock.
Access to the everyday default model with tight limits, a small context window that silently truncates long documents, and - as of early 2026 - ads. Enough to explore, not enough to build. During peak times, you may wait in line.
The budget middle tier: more volume than free, still ad-supported. The catch that reviewers keep flagging: it does not include the flagship model. If you need more than free but aren't doing professional work, Go exists. Most working creatives should skip straight past it.
The workhorse, held at twenty dollars for three years. The flagship model, ad-free, image generation, Deep Research runs, Sora video, the Codex coding agent, Agent Mode, and custom GPTs. For most professionals, Plus is the plan that makes ChatGPT useful for real work.
Two rungs, same idea: capacity. The $100 tier runs about five times Plus with the top reasoning variant. The $200 tier runs about twenty times, with unlimited video generation and the Operator agent. Steep - but for people who live inside the tool all day, it removes every friction point. You're buying headroom, not new features.
Google bundles Gemini into its broader ecosystem, which makes pricing a little different. The paid tiers come packaged with serious cloud storage and Workspace integration - so you're not just buying an AI, you're buying deeper access to Google's entire world. One naming note: these plans have been renamed more than once, so older articles quote dead names and dead prices. Trust the current menu, not the archive.
Gemini app access with a handful of Deep Research reports a month and basic storage. Good for quick questions and light research. Tighter caps than the paid competitors.
The cheapest paid seat among all the majors, and unlike ChatGPT's budget rung, a real option: more model access plus hundreds of gigabytes of storage. If five dollars is the budget, this is where it goes.
The flagship model with a million-token context window - the cheapest long-context seat in the market - plus Deep Research, Veo video generation, Gemini inside Gmail and Docs, terabytes of storage, and a YouTube Premium Lite bundle. If you already live in Google's ecosystem, this is where the AI becomes part of your workflow instead of a separate tab.
The power tiers: five-times and twenty-times usage, deeper reasoning modes, an always-on agent, massive AI credit pools for video and image generation, and tens of terabytes of storage. Overkill for most writers, but worth knowing it exists.
Copilot's pricing is the most tangled of the Big Four because it's layered on top of Microsoft 365 subscriptions. The AI isn't sold alone anymore - it rides along with Word, Excel, Outlook, and the rest. For individuals, the key question is whether you already pay for Microsoft 365.
The consumer Copilot chat app. Useful for quick questions, but no integration with Office apps. Think of it as a chatbot that happens to live in your browser.
Gets you Copilot inside Word, Excel, PowerPoint, and Outlook with a monthly AI credit allowance. Not unlimited, but enough for moderate use. The AI shows up where you already work.
The individual power tier, and the plan that absorbed the old standalone Copilot Pro when it was retired. Uncapped-tier Copilot across the Office apps, priority model access, the full desktop suite, and terabytes of OneDrive. If your deliverables live in Office, this is the seat.
Business Copilot runs $18 to $30 per seat as an add-on - and note the word add-on: it requires a qualifying Microsoft 365 license underneath, the hidden cost most guides skip. A different conversation entirely.
Quick reference for the tools we covered in Module 3. None of these are essential for this guide, but if one caught your eye, here's what the door costs.
The free tier handles basic research queries with citations. Pro unlocks Deep Research, higher-powered models, unlimited searches - and the quiet killer feature at this price: a model switcher that puts the current GPT, Claude, and Gemini flagships inside one seat. If you want an AI that shows its sources, twenty bucks gets you there.
Access to Grok comes through X subscriptions, and the packaging has shifted more than once - check the current menu before budgeting for it. The AI is capable, but the entry point is through a social platform, which may or may not fit your workflow.
No paid tier as of this writing. Pi is designed as a conversational companion, not a production tool. If you want to think out loud without a meter running, it's there - with the same caution from Module 3: confirm it still exists in the form described.
The free version lets you sample multiple models with daily limits. Paid tiers unlock higher usage across the full model lineup. Useful for comparison shopping - seeing how different AIs handle the same prompt.
There's a feature in Claude that most people don't know about until they see it happen. It's called conversation compaction, and it's one of the reasons Claude handles long creative sessions better than anything else on the market.
Here's how it works. Every conversation has a maximum length - a limit on how much text the AI can hold in its active memory. When your conversation starts approaching that ceiling, Claude automatically compresses the earlier parts of the thread. It distills the key information, preserves the working context, and makes room for the conversation to keep going.
Important: you cannot trigger this yourself. Don't type "hey, can you compress this?" - it doesn't work that way. Compaction is an internal process Claude manages on its own, in the background, when the thread needs more room. You won't see it happen. You'll just notice the conversation keeps going when you expected it to stop.
Now the honest part, because it matters for how you work. Compression means distillation, and distillation has a cost: the broad strokes survive, but the finest details from a hundred pages back can fade. A thread that's been compacted several times remembers the shape of every decision - not always the exact wording of each one. Claude will compress a conversation multiple times before it finally maxes out. In the early days I copied finished threads into Google Docs that ran three, four, even five hundred pages. That's the real scale of what a long build holds.
Long conversations now load selectively: the app keeps what's near your position, not the entire scroll. Select-all grabs the neighborhood, and a long session exports truncated or scrambled.
So capture as you go: pull the passages that matter while you're near them, and file decisions the moment they prove themselves. Wholesale rescue at the end is no longer a promise.
That's what the habits in the later modules are for: keeping a running summary, saving choices while they're fresh, building a vault that holds the whole project. The thread carries you far. It's the source-record of your creation. Put a system around those records and they'll carry your operations forever.
One more thing to know as you choose your tier. These platforms - especially Claude and ChatGPT - have project features, memory systems, and workspace tools that turn your AI into more than a conversation partner. Over time, it becomes a container for your work. A vault.
We'll dig deep into those features in later modules. For now, just know that the paid tiers don't just give you more messages. They give you a place to build something that persists. And that changes everything about how you work.
Every conversation about AI pricing focuses on what you get. More messages. Faster responses. Better models. But the comparison charts never list what changes in you.
When the tool can keep up - when you're not rationing your ideas or cutting a session short because you're watching a usage meter - something shifts. You stop editing yourself before you even start. You think bigger. You go longer. You follow the weird tangent that turns out to be the whole point.
That's not a feature on a pricing page. That's what happens when friction disappears.