Three kinds of "free" — learn to tell them apart
Before the list, a filter. In practice, "free" AI tools fall into three groups, and only the first two belong in your daily workflow.
Genuinely free and open. The model runs on your own hardware, or the weights are openly published. Nobody is paying per token because there are no tokens. Whisper, Ollama and LM Studio live here. No cap, no card, no server watching what you type.
Freemium with honest limits. A company hands you a real slice of its paid product — a few thousand characters a month, a handful of images a day — because it wants you as a future customer. This is where most of the genuinely useful tools sit.
"Free" that bills your data. No money changes hands, but your prompts and uploads may train future models. That is not automatically a bad deal. It is a price, and it should be a conscious one.
The assistant layer: five free chatbots, one of them European
The big general assistants all have real free tiers in 2026, and the differences between them matter less than the settings inside them.
ChatGPT still gives the most forgiving free tier for everyday drafting, explaining and translating, with a daily cap on the strongest model. Google Gemini is the natural pick if you already live in Gmail and Docs, and it links answers to web sources. Claude is the one to reach for when you need to read a long document carefully rather than generate volume. Microsoft Copilot is worth having on Windows simply because it is already there.
Then there is the European option. Mistral's Le Chat, from the Paris-based lab, has a free tier that answers almost instantly and handles European languages — Czech, Polish, German, French — noticeably better than some American models do. For a European reader, that alone justifies a bookmark.
Writing, translation and research
DeepL remains the most reliable free translator for European language pairs, and it is a German company, which simplifies the compliance conversation for many organisations. The free web and app versions need no card; the paid tiers add glossaries and document handling.
NotebookLM is the underrated one. You upload your own PDFs, lecture notes or reports, and it answers only from those sources — which sharply reduces the invented-citation problem that plagues free chatbots. For students and anyone drowning in documentation, it is the single most useful free tool on this page.
Perplexity rounds out the set with source-linked answers; the free tier simply caps how many of the stronger searches you get per day.
Coding, even if you are not a developer
GitHub Copilot Free gives individual accounts a monthly allowance of code completions and chat requests without a card — enough to be genuinely useful if you write scripts occasionally. Google AI Studio is free with rate limits and lets you experiment with current Gemini models, including longer context windows, before you commit to anything.
Images, voice and video
Adobe Firefly hands out a monthly allowance of generative credits to anyone with a free Adobe account, and Adobe positions it as trained on licensed and Adobe Stock material — a meaningful difference if you intend to publish the result. Canva bundles AI image and text tools into its free plan, which is usually enough for social posts and slides.
For voice, ElevenLabs gives a modest monthly character allowance that covers a short voiceover or a few podcast intros, typically with attribution required. For music sketches, Suno's free tier generates a small number of tracks per day, though commercial use generally requires a paid plan — read that clause before you publish anything.
One caution worth stating plainly: CapCut, the free editor many people now rely on, is owned by ByteDance. That does not make it unusable, but if you handle client footage, check your contractual obligations before uploading it to a free consumer service.
The genuinely free option: run it on your own machine
If the content is sensitive — a patient's notes, a student's essay, an unpublished client brief — the safest free tool is the one that never leaves your laptop. Ollama and LM Studio both let you run open-weight models such as Llama, Mistral, Gemma or Qwen locally, with no account and no card. Whisper, OpenAI's openly published speech model, does transcription offline with excellent accuracy on European languages.
The trade-off is hardware. A machine with 16 GB of RAM will run a small model comfortably; the results will not match a frontier cloud model, and you will feel it on longer tasks. Our own local-model testing on the AI Arena rig is a reasonable place to start if you want to know what your own hardware can realistically handle.
Free tiers compared
| Tool | Best for | Card needed? | What to check for EU use |
|---|---|---|---|
| ChatGPT (free) | Everyday drafting and questions | No | Training opt-out toggle in data settings |
| Google Gemini | Search-linked answers, Workspace users | No | Consumer and Workspace data terms differ sharply |
| Claude (free) | Long documents, careful writing | No | Daily message cap; read the current data-use notice |
| Mistral Le Chat | European languages, fast answers | No | French provider, EU-based |
| DeepL | Translation across European languages | No | German company; free tier is text-limited |
| NotebookLM | Summarising your own documents | No | Answers grounded in your uploads |
| Perplexity | Source-linked research | No | Only source-linked answers should be reused |
| Adobe Firefly | Images for commercial projects | No | Monthly credit allowance; provenance metadata |
| ElevenLabs | Voiceover for video and podcasts | No | Attribution usually required on free tier |
| Ollama / LM Studio | Fully offline work | No | Nothing leaves your machine |
Quotas move constantly. Treat these descriptions as a starting point and check the vendor's own page before you build a workflow on one.
The European angle: free still means GDPR
Here is the part that catches people out. A free tool is not exempt from the GDPR just because you paid nothing. If a provider processes personal data of people in the EU, the rules apply — and fines still reach up to €20 million or 4% of global annual turnover, whichever is higher.
The practical consequence is narrow but important. Under Article 28, using a provider to process personal data on your organisation's behalf requires a data processing agreement. Consumer free tiers almost never offer one; business tiers usually do. That single fact decides whether a teacher can paste a pupil's essay into a free chatbot (no) or a small business can run its own public marketing copy through one (usually yes, if no personal data is involved).
The EU AI Act adds a second layer. Prohibited practices and AI-literacy duties have applied since February 2025, and the general-purpose AI obligations — including publishing summaries of training data and respecting copyright opt-outs — since August 2025. The bulk of the remaining rules, including the transparency duties that require chatbots to tell you that you are talking to a machine, apply from 2 August 2026. In plain language: from this year, a free chatbot in Europe should not be able to pass itself off as a human.
A five-minute checklist before you sign up
- Find the training toggle first. Open Settings → Data controls before you type anything worth keeping. If there is no opt-out, assume your text may train a model.
- Check where processing happens. Some providers offer an EU data region; many do not. This matters more than the price.
- Read the free-tier table, not the landing page. The limit you hit in week three is always documented — just not on the homepage.
- Never paste third-party personal data — pupils, patients, clients — into a consumer free tier without a legal basis and a signed agreement.
- When in doubt, go local. A smaller model on your own machine beats a bigger one you cannot legally use.
Are free AI tools safe to use for work?
It depends on the data, not the price. Public marketing copy or a rough draft is usually fine. Anything containing personal data about other people generally is not, because consumer free tiers rarely come with the data processing agreement that GDPR Article 28 requires. When in doubt, ask your employer which tools are approved — many now maintain a list.
Will these free tiers still exist next year?
No guarantees. Free tiers exist to convert users, and providers adjust them regularly — sometimes upward, often downward. The safest long-term strategy is to avoid building a critical workflow on one vendor's free allowance, and to keep an offline option ready for sensitive work.
Do I have to tell people when I use AI?
Increasingly, yes. From 2 August 2026, the EU AI Act's transparency rules require that people be informed when they are interacting with an AI system, and synthetic or manipulated media must be labelled. Separately, schools and publishers usually have their own disclosure rules that apply regardless of the law.
None of this is about avoiding AI. It is about knowing which free tool is paying for itself and which one is being paid for by you. Pick two from the list above, find the training toggle in both, and you will already be ahead of most people who clicked "start free trial" this morning.
