I Spent a Week Using Only Free AI Tools. Here Is What Happened.
So I cancelled ChatGPT Plus, Claude Pro, and Midjourney, all in the same week, and lived entirely on free tiers for seven days straight. Not because I'm broke. Because I wanted to know if the free version of the AI boom is actually usable, or if it's a demo built to convert you within 48 hours.
Day one: fine. Suspiciously fine. ChatGPT's free tier still runs on GPT-4o, the same model Plus subscribers get, just on a leash. According to testing published by PE Collective earlier this year, free users get somewhere around 15 to 40 GPT-4o messages per three-hour window before getting quietly bumped down to GPT-4o mini. Nobody tells you when that happens. You just notice the answers get dumber.
Day two, still coasting. Claude doesn't run free users through a stripped-down model the way ChatGPT does. Anthropic hands out actual Sonnet, not some mini variant, which is generous on quality and stingy on quantity. You get real capability, then you hit a wall fast. Sometimes after a single long conversation.
Day three is where it got ugly. Mid coding session, three files deep, the kind of task where the AI is the only thing keeping the mental model in your head from collapsing, and ChatGPT cut me off. Switched to Claude. Hit its cap twenty minutes later. Finished the session in a Discord server running someone's self-hosted Llama endpoint, which is not a sentence I expected to type in 2026.
Day four, I found the actual MVP of free-tier AI, and it wasn't one of the household names. Meta AI, the assistant baked into Instagram and WhatsApp, doesn't throttle you the way the standalone chatbot apps do. It's not the smartest model in the room. It's the one that's actually there at 11pm, on your fourth attempt at a cover letter, without a paywall sliding into frame. HuggingChat picked up anything more experimental. Open-ended brainstorming, the kind of task where a slightly dumber model with zero rate limit beats a genius model that logs you out mid-thought.
By day six I had an actual system. Not because I planned one. Scarcity forces triage. Quick factual lookups went to Meta AI. Anything requiring real reasoning got rationed to Claude, one careful prompt at a time, no wasted follow-ups. Coding went wherever had quota left, which some days meant three chat windows open for a single afternoon of work.
Day seven, verdict: you can do it. A full week of real knowledge work on free-tier AI alone, no subscriptions. But it's death by a thousand paper cuts, not one wall you hit and give up on. The free tier isn't broken. It's engineered. Every rate limit sits exactly where it needs to sit to get you hooked on the good version before yanking it away at the worst possible moment. Three files deep. Mid-thought. Out of quota.
This isn't a new trick. It's the Spotify free-tier-with-ads model and the mobile-game energy-system model, run through a chatbot instead of a music app or a phone game. The AI industry didn't invent the freemium funnel. It inherited it, fully formed, from twenty years of SaaS psychology, then bolted it onto a product expensive enough that the incentive to convert you is sharper than usual.
What's actually new is how fast the math flips. A free trial for a photo-editing app costs a company almost nothing to run. A free GPT-4o conversation costs real inference compute, every message, which is exactly why the frontier models ration you so much harder than something like Meta AI's more modest assistant does. The generosity of a free tier is basically a readout of what that specific model costs to serve per query. Cheap model, long leash. Frontier model, short one. You can reverse-engineer a company's cost structure just by watching where it throttles you.
Would I do it again by choice? Probably not. But the free tier is a real option, not just a taste, if you're patient enough to juggle three apps and rebuild your workflow around scarcity instead of convenience. Just don't expect anyone to tell you, mid-conversation, that the model quietly got worse. You have to notice that part yourself. NotebookLM turned out to be the one free tool on the list that never once made me feel like I was settling, which says something about which parts of this industry are actually optimizing for usefulness instead of conversion.
Enjoyed this? Get more.
Weekly dispatches on AI culture, chatbots, and the robot future. No hype.