Free plans change often
Free AI plans are rewritten more often than almost any other product page. Models get swapped, features move between plans and limits are tuned without notice. That is why every row in the table above shows the date we opened the vendor's page and links straight to it. If a row matters for a decision, open the source and read today's version before you rely on it.
We copied only what each page says. When a page names no number, we name none either. When a page does not mention a feature for the free plan, the table does not claim it.
How to compare free plans
Vendors describe their free plans in different words, so compare the same few points for each one:
| Check | What to look for on the pricing page |
|---|---|
| Model | Which model the free plan uses, and which models are paid only |
| Limits | Whether the page gives a number, a time window or only says "limited" |
| Features | Web search, image generation, file uploads and research modes named for the free plan |
| Account | Whether you need an account, and which one |
| Platforms | Web, phone apps, desktop apps |
Then test with your own work. Paste the same real task into two or three assistants and compare the answers. A clear prompt makes the test fair: the prompt builder puts role, task, context and format in one block you can reuse, and the token counter shows how long your document is before you paste it.
Why limits move
Running large models costs money for every message, so free limits follow demand and the vendor's capacity. Most pages say this openly instead of giving a fixed count. Microsoft describes a limited free version "subject to available capacity and limits" on its plans page. Anthropic's pricing page says free limits apply over a rolling five-hour window, and Google's subscriptions page says limits refresh every five hours until you reach a weekly limit.
In practice: long chats and big files use up a limit faster than short questions, and a limit you hit today may be different next month. Start a new chat for a new topic and keep only what the assistant needs in the prompt.
Personal and confidential data in chats
Treat a free chat like an email to a company you do not know well. Leave out passwords, customer data, health details and anything under a confidentiality agreement, or replace names and numbers with placeholders before you paste.
The vendors' own pages show why this matters. Google's Gemini privacy hub says human reviewers look at some chats and asks you not to enter confidential information you would not want a reviewer to see. Anthropic's privacy policy says it may use your chats to train its models unless you opt out in your account settings. DuckDuckGo's Duck.ai privacy page says it removes your IP address before a prompt reaches the model provider and that the providers agree not to train on those chats.
Before you use any assistant for work, read its privacy page and check the training and history settings in your account.
Tools for working with AI
A few free tools on this site help around the chat itself:
- Token counter: how many tokens a prompt or document has.
- AI cost comparison and AI token cost calculator: what the same job costs through the paid APIs, once a free plan is not enough.
- AI text cleaner: removes markdown, hidden characters and odd spaces from text you copy out of a chat.
More free templates and downloads are on the freebies page.
How we checked
We opened each vendor's own pricing page or help center article on the date shown in the table and wrote down only what it says about the free plan. No roundups, no reviews, no numbers from memory. Entries whose official page we could not read were left out rather than filled from other sources. The filter chips only mark a feature when the source names it for the free plan, so a missing chip means "not stated on the page", not "not available".
Frequently asked questions
Which AI chatbot has the best free plan?
It depends on the job. For answers with web sources, a search focused assistant fits; for images, pick one whose free plan names image generation; for privacy, look at what the vendor says about training and storage. Test two or three with the same real task, and check the date on each row, because free plans change often.
Can I use an AI chatbot without an account?
Some work for a first try. Google's Gemini help says some features of the web app work without signing in, and Perplexity's help center says it can be used without signing in, with some but not all features. Saved chats and the extra features need an account. Our list of AI tools without signup has more.
Which free plans include image generation?
On the day we checked, the free plan pages of Gemini, Vibe by Mistral and Meta AI named image generation or creation, mostly with lower limits than paid plans. Perplexity's help center says its free plan has no image generation.
Why do I hit the free limit so quickly?
Free limits depend on the vendor's capacity and are often counted over a time window rather than per day. Long chats, big files and research modes use more of the limit than short questions. Starting a new chat for each topic and pasting only what is needed helps the limit last longer.
Do free plans use the newest models?
Not always. On the day we checked, Claude's free plan listed Sonnet and Haiku but not Opus, Gemini's free plan named the Flash-Lite model, and Perplexity's free plan had no access to its advanced models. The table names the models each free plan page mentions.
Are my chats used to train the AI?
It depends on the vendor and your settings. Anthropic says it may train on chats unless you opt out in your account settings, Google says human reviewers see some Gemini chats unless you turn off Keep Activity, and DuckDuckGo says its model providers agree not to train on Duck.ai chats. Read each privacy page before you paste anything sensitive.