Practical guide · verified against the real thing
Free AI tools worth using (and where the free tier actually ends)
In one line: Every major AI assistant has a free tier — and each one ends differently. How the caps behave, what they're good for, and how to test where yours runs out.
The free tiers of the major AI assistants have quietly become daily-driver tools — and each one runs out in a different way. This page maps the shapes honestly: what each free tier is good for, how its limits behave (no invented numbers — the providers don't publish fixed ones, and third-party counts conflict), and the working strategy for living free. Checked September 2026; this page has the shortest expiry of anything on the desk.
ChatGPT free: the metered everything-store
OpenAI's free tier gives limited access to the current flagship model with the full app experience around it — voice, image analysis, the works — but flagship messages are metered on a rolling window, and when the meter runs out, the assistant quietly hands you to a smaller, faster model until the window refreshes. You're never locked out; you're downgraded. Behaviour to expect: a burst of excellent answers, then subtler ones. Best free use: varied everyday tasks, voice, and trying the ecosystem before paying (pricing if you do: $8 Go, $20 Plus).
Claude free: the working writer's window
Anthropic's free tier runs its flagship-class models on rolling usage windows that flex with demand — busy hours burn the allowance faster, and when it's gone, it's gone until the window resets. No smaller-model fallback; just a polite wall. Best free use: long-form writing, long documents (Claude handles big files natively), and the draft-and-revise loops where its prose needs the least editing. The honest catch: a working afternoon can spend a free tier by lunch, which is why the $20 Pro tier is the one free users graduate to first.
Gemini free: the generous generalist
Google's free tier runs its Flash-class model with the most forgiving limits of the three — no published cap, throttling rather than walls, plus the integration superpower: it lives inside Gmail, Docs, Drive and Search where your material already is. Trade-offs: the free model is a class below the paid flagship for hard reasoning, and the integration is also the data relationship (review the training settings — we did).
The supporting cast, briefly
Perplexity — search-style answers with citations; free tier covers unlimited basic searches with a small daily quota of its deeper Pro mode: the free research tool of record. Microsoft Copilot — the GPT-class assistant with web grounding, free, woven into Windows and Bing; caps exist, throttles politely. DeepSeek — the free-and-cheap challenger whose consumer app is genuinely free (the value case, examined); treat data-sharing assumptions accordingly. And the non-chatbot free AI that's quietly excellent: the models inside Google Docs' writing help and Notion's AI tier — free enough to try, billed enough to notice (Notion's free line, exactly drawn).
The strategy that works
Don't marry one free tier — rotate by task: Claude free for the writing session, ChatGPT free for the everything-else afternoon, Gemini free for anything that lives in Google, Perplexity for research with receipts. The rotation isn't a hack; it's what the pricing actually models — each provider is subsidising a different kind of试用 usage. And the honest end-of-tier ritual: when you hit the same wall three days running, that's the market telling you which subscription your actual usage wants. Buy the one you ran out of, not the one with the best commercial (and keep the hype test handy).
Checked September 2026 · free-tier limits are deliberately unpublished and change without notice — the behaviours above are the durable part; numbers are not.
Sources
Next