Every paid AI tool now has a free cousin. Some of those cousins are nearly as capable. Others are demo shells that hit a paywall twenty prompts in, usually mid-task and always at the worst moment. The difference between the two has almost nothing to do with how smart the underlying model is and almost everything to do with how the free tier is metered.
Over the past few months I’ve run the same eight jobs through paid and free versions side by side: a 2,000-word draft, a competitor teardown, a 45-minute call transcript, a stubborn Python bug, a budget spreadsheet, a résumé rewrite, a product mockup image, and a research question that needed real citations. Same prompts, same files. Here is where the free alternatives genuinely hold up, and the specific points where they quietly stop.
Chat assistants: three free tiers cover most text work
The free versions of ChatGPT, Claude and Gemini do roughly 80% of what most people pay for. The caps reset every few hours, so a morning of drafting and editing is usually fine. Where they bite is deep in a long session, exactly when you’ve fed in a 30-page pdf and the tool tells you to come back later.
Practical differences matter more than benchmark scores. ChatGPT’s free tier gives you the current model with a message cap, plus limited image generation and a few deep-research runs per month. Claude’s free plan hands you a strong model with a shorter message window, which is fine for careful editing and miserable if you’re brainstorming 40 taglines in a row. Gemini’s free tier is the most generous on raw volume, though the default model is the faster, lighter one unless you accept tighter limits on the Pro model.
If you want the full breakdown of who wins on what, my piece on how Claude stacks up against ChatGPT, Gemini and the rest goes tool by tool, including the tasks where the cheaper option is the better one. For a broader shortlist, this roundup of free AI chatbots that are worth your time filters out the ones that are just a branded wrapper around someone else’s API.
Research and search: where free is arguably better
For cited answers, you don’t need a subscription at all. Perplexity’s free tier gives you a set number of Pro searches per day (currently around five) and unlimited basic searches, which are perfectly usable. ChatGPT’s search mode works without paying. Google’s AI Mode is free and increasingly hard to tell apart from a good research assistant.
The interesting outlier is Andi, which runs ad-free and shows its sources inline rather than burying them under a summary. It’s slower and less polished than Perplexity, and its index is narrower, but the absence of sponsored results changes the feel of the output. Worth reading a direct comparison before you commit your muscle memory to one of them: Andi Search versus Perplexity, Google and ChatGPT puts the four side by side on the same queries.
Meeting notes: the free tiers that actually record
Transcription is the category where free plans are most deceptive. Many tools advertise a free tier that turns out to be a seven-day trial. Fathom is the exception I keep coming back to: unlimited recording and transcription at no cost, with AI summaries limited to a handful of meetings per month. For someone who needs the transcript more than the summary, that’s close to a full product.
Otter’s free plan gives you 300 minutes a month and caps each conversation at 30 minutes, which is fine for interviews and useless for a two-hour workshop. Granola and similar Mac-first apps are generous in beta but change terms often, so don’t build a workflow on them without a backup. If you want a realistic picture of the setup, this step-by-step Fathom workflow from install to follow-up email shows how far the free tier stretches on a normal week of calls.
Images, audio and video: free, with watermarks and quotas
Free image generation is good enough for mockups, blog headers and moodboards, as long as you accept a few limits. Expect daily token caps, slow queues at busy hours, and commercial-use terms that vary wildly between tools.
- Ideogram — free daily generations, best-in-class at putting legible words inside an image, slower at peak times.
- Leonardo and Krea — a fixed daily allowance of credits that refills, good for consistent styles across a set.
- Copilot / Designer — free and fast, though the output has a recognisable house look.
- ElevenLabs — 10,000 characters a month of text-to-speech, which is roughly 10 minutes of audio, enough to test a voiceover idea.
- Adobe Podcast Enhance — free cleanup that rescues bad room audio better than most paid plugins.
- CapCut — free editing with auto-captions; the paid features are mostly templates and export options.
For video, free means rationed. Runway hands out a one-time credit pool, Kling and Pika give daily generations, and all of them take far longer than the paid queue. Ten seconds of usable footage is a realistic ceiling in a free session.
Coding assistants: the most generous free tiers of any category
GitHub Copilot’s free plan gives 2,000 code completions and 50 chat requests a month, which sounds stingy until you realise most people only hit chat a few times a day. Gemini Code Assist for individuals is far looser, with a very high monthly completion allowance. Cursor and Windsurf both run free tiers with limited premium requests, and Codeium’s extension remains free for personal use.
The catch is context. Free coding tools tend to truncate your repository, so they’re sharp on a single file and lost inside a monorepo. If your bug lives in one function, free is plenty. If it lives in the interaction between four services, you’ll want the paid window.
The hidden costs nobody puts on the pricing page
Free tiers are not free in the ways that matter six months in. Four things to check before you build a habit:
- Data use. Several free plans reserve the right to train on your inputs. Read that clause before you paste in a client contract.
- Commercial rights. Some free image tools grant you a licence only for personal projects, or require attribution.
- Model downgrades. Free users are usually routed to a cheaper model at peak times, so the same prompt can behave differently at 9am and 9pm.
- Term changes. Generous free tiers get cut regularly. Anything critical needs a fallback that isn’t the same company.
How I’d build a genuinely $0 stack
Pick one primary chat assistant and learn its quirks instead of spreading yourself across four. Use Perplexity or Andi for anything that needs a source you can click. Let Fathom record everything and skip the AI summary on most calls. Generate images in Ideogram, edit audio in Adobe Podcast Enhance, and keep Copilot in the editor for the small stuff.
Then fill the gaps with something narrower. For thinking out loud, a conversational assistant such as Pi is a better fit than a general chatbot because it asks questions instead of dumping a list; the comparison of Pi against ChatGPT, Claude, Gemini and Perplexity explains where that difference actually shows up in use.
Test any free tool in ten minutes before you commit
Don’t evaluate a free tier with a throwaway prompt. Take a real task you finished last week, feed it the same source material you used then, and watch three things: how many prompts it takes before the tool gets useful, whether it remembers the constraints you set five messages ago, and what happens when you ask it to change one specific sentence rather than rewrite everything.
Then do the boring check. Export the output. If a tool won’t let you take your work out in a usable format without upgrading, the free tier is a trap, not a trial. The ones that pass all four tests are worth keeping in rotation; the rest are worth deleting from your bookmarks so they stop wasting your afternoons.