You can write an email, summarise a page, make a picture, translate a paragraph and plan a budget without paying anything. What catches people out is that each app stops you at a different point, and the free versions don’t all measure your usage the same way. The limits changed quickly in 2026, so a guide written in spring may already be wrong about the free AI tools you use.
The biggest recent change is ChatGPT’s. It now gives free users unlimited plain-text chat, which leaves many people thinking every other feature is unlimited too. It isn’t. Every limit below was checked against the companies’ own help pages and plan tables on 29/09/2026, and those pages change often. Once you know what each free plan actually counts, you can pick a tool that won’t stop you halfway through a task.
The 2026 Free-Tier Reality: Message Caps, Credits, and Context Windows
A “free tier” in late 2026 isn’t one kind of limit. Each company meters something different, so comparing them by counting messages gives you the wrong picture.
ChatGPT removed the message cap on text chat. Since 06/08/2026, text chats have been uncapped on every ChatGPT plan, and free accounts run on a model called GPT-5.6 Luna. OpenAI’s own announcement said separate limits remain for files, images, voice and image generation.
Gemini measures effort rather than messages. Google says its limits are compute-based and “factor in the complexity of your prompt, the models and features you use, and the length of your chat”. The allowance refreshes every 5 hours until you reach a weekly limit. In plain terms, a long document or a picture uses up more of your allowance than a short question does.
Claude works to a time window. Anthropic doesn’t publish a message count for its free plan. It says usage depends on the length and complexity of your conversations, the features you use and the model you’re chatting with. Its Pro plan gives at least five times the free plan’s usage per session, with the session resetting every five hours.
Perplexity rations its deeper searches. Its plan table lists 3 Pro Searches a day and 1 Research query a month on the free plan, with “practically unlimited” basic searches. Some guides still say five Pro Searches a day. Perplexity’s own table now says three.
There is one more limit to know about, the context window. This is how much text the AI can “read” at one time, and it decides whether a long document fits at all. Google publishes this for Gemini: 32k tokens without a paid plan, 128k on AI Plus and 1 million on AI Pro. A token is a small piece of a word.
| Tool (free plan) | What it counts | Published free allowance, as of 29/09/2026 | When it comes back |
|---|---|---|---|
| ChatGPT | Files, images, voice (text is uncapped) | No numbers published for files or images; 500MB Library storage | Rolling windows that start from first use, not midnight |
| Claude | Overall usage per session | No number published | Five-hour session |
| Gemini | Compute (effort per prompt) | “Standard limits”; 32k-token context window | Every 5 hours, inside a weekly cap |
| Perplexity | Pro Searches and Research | 3 Pro Searches/day, 1 Research/month | Daily / monthly |
| Microsoft Copilot | Chat, images, uploads | Microsoft says only “subject to available capacity and limits” | Microsoft doesn’t say |
These tools also offer different privacy settings. In ChatGPT, go to Settings, then Data controls, and turn off “Improve the model for everyone”. OpenAI says your new conversations then won’t be used to train its models. In Claude, the switch is under Settings, then Privacy. In Gemini, turning off Keep Activity stops future chats being used for training, but Google still keeps them for 72 hours.
Writing and Planning: Navigating Text and Chat Limits
Writing emails and planning a budget are mostly typing, and typing is where the free plans are most generous. The limits appear once you attach files or ask for heavier features in the same conversation.
ChatGPT is the easiest free option for long writing sessions right now, because plain text has no message count. OpenAI lists the costs. In the United States, free users have seen labelled ads under some answers since 09/02/2026. Whether those ads have reached Nigeria couldn’t be confirmed for this article. Free users also get a Think button that gives the model more time on a hard question, like checking a monthly budget. OpenAI puts guardrails against abuse on heavy use of it. The newest model, GPT-6 Astra, isn’t available on Free or Go at all, according to a tracker citing OpenAI’s help pages.
Claude works differently. Every message uses part of your five-hour allowance, and long pastes use more. Anthropic says longer conversations that trigger automatic context management consume more of your usage, and suggests starting a new chat when you’re close to the limit. If you plan to draft a cover letter and then rework it ten times, one detailed prompt will get you further than many small follow-ups. Anthropic’s advice to batch related requests into one message points the same way. Its help centre also describes progress bars under Settings, then Usage that show how much of your allowance is left.
Students using these free AI tools should know about watermarking. In August 2026, TechCrunch reported on Anthropic’s move to include a watermark in text and files generated by Claude to comply with EU rules. OpenAI’s terms say you must not represent that output was human-generated when it was not. A free chatbot can help you plan and edit an essay, but passing off its text as your own breaks the provider’s rules. Some watermarks exist to detect exactly that.
Budgets and similar money tasks come with their own warning. OpenAI says you should not rely on output as a substitute for professional advice. Check any figures the AI adds up against your own bank statement before you act on them.
Summarising and Translating: File-Size Walls and Character Counts
Summarising a pasted paragraph rarely hits a limit. Uploading a file is what triggers them, and each tool counts files differently.
Gemini’s 32k-token context window is the main limit for long documents on the free plan. A short report will fit, but a whole textbook chapter may not. Gemini won’t show you a number going down. Google says only that media generation and Deep Research consume more of your usage, and uploaded files count as “complexity” in the same way.
Perplexity is good at summarising a web page because it cites its sources, but its free plan offers only basic file uploads, marked “Limited”, with no published number. An aisparkhub reviewer who tested the free plan on 02/09/2026 wrote that Perplexity sometimes misunderstood uploaded material and that they had to check its answers against the original sources. That’s one person’s experience, not a measured error rate, but it’s worth copying the habit: read the cited source before you trust a summary.
ChatGPT’s free file uploads are capped separately from its text chat. Uploaded documents also count towards a 500MB Library limit on free accounts.
For translation, DeepL publishes the clearest free limits. Its plan comparison lists 50,000 characters a month and 1 file translation of up to 5 MB on the free version. Some third-party sites quote 500,000 characters, but that figure is for DeepL’s separate API Free plan for developers, not the translator you use in a browser. As a rough guide, one eesel.ai writer estimated 50,000 characters at about 8,300 words a month. That’s plenty for a few letters but not enough for a thesis. If you need one long file translated, DeepL’s single monthly file may be the whole of your free allowance, so it may be better to paste the text in sections instead.
Generating Images: Daily Boosts, Watermarks, and Usage Rights
Images use far more of a free allowance than anything else, and the limits here are the least openly published.
Microsoft Copilot is the tool usually recommended for free images, because of its “boosts”, which are credits for fast image generation. Guides written before Microsoft’s app update, such as copilot-experts.com in July 2026, said that Microsoft documents 15 image-generation boosts per day for free users. Copilot moved to an updated app on 18/08/2026, and Microsoft’s page now says only that you can create images for free, “subject to available capacity and limits”. It adds that when you hit a limit you can upgrade or come back later. Treat 15 as a pre-update figure rather than a guarantee.
The lack of a visible number frustrates people. One Copilot user on Microsoft Q&A wrote that they kept getting the image-limit message but couldn’t find out what the limit was or when it would reset. The Microsoft Q&A reply described the system as dynamic and suggested waiting at least 24 hours after the last image.
Gemini includes image generation on its free plan, and media generation consumes more of your usage than chat. A day of making pictures can therefore leave less for text. ChatGPT’s free plan keeps its own cap on image creation, separate from text. Perplexity’s free plan doesn’t include image generation at all.
On watermarks, Google announced in August 2026 that users can switch off the visible Gemini watermark under Settings, then Media Watermark, with the change rolling out in the following days. The invisible SynthID watermark and C2PA metadata stay in the image. C2PA metadata is hidden information in the file that records where it came from. OpenAI says images from ChatGPT carry both C2PA metadata and SynthID watermarks. Hiding the visible mark doesn’t make an image untraceable.
Usage rights matter if you sell anything. OpenAI’s terms say that, between you and OpenAI, you own the output. Copilot’s consumer terms say you can only use Copilot for your own personal use and forbid removing the content credentials that show Copilot made something. That makes free Copilot images a poor choice for a business flyer, though fine for a birthday card. This isn’t legal advice, so read the current terms yourself before using an image commercially.
The “No Sign-Up” Route: When You Just Need One Quick Answer
Some tasks don’t justify creating another account. There are still a few ways to use AI without one, but they’re narrower than many guides suggest.
Duck.ai, from DuckDuckGo, is the most clearly documented option. Its site says no account is required, and DuckDuckGo says that by default it doesn’t record or store your chats, and they aren’t used to train models by DuckDuckGo or the model providers behind it. The free version has a daily limit that resets at midnight UTC. Nigeria runs on West Africa Time, one hour ahead of UTC, so the reset lands at 1:00 a.m. local time. DuckDuckGo also warns that you may see an error when using a VPN or a cellular network, and suggests switching to Wi-Fi.
ChatGPT also works without signing in. OpenAI’s help pages describe signed-out use, where your training setting is saved in that browser. You lose saved history and the account features. Gemini supports signed-out use as well: Google says such data is handled under its general privacy policy and not linked to a Google account. A 2025 9to5Google report added that uploading files still required signing in and that the Android app still needed a login.
Other guest routes are shrinking. Copilot’s terms state that to use Copilot you must have a Microsoft Account. One user on Microsoft’s Tech Community reported that guest access in Edge InPrivate on Android now shows a sign-in screen with no way to skip it. Perplexity still advertises logged-out basic search. However, one Reddit user described in March 2026 how a login overlay eventually blocked every answer on their browser. These are individual reports, not official policy, but they show that the guest route can close without warning.
Decision Tree: Match Your Task to the Right Free Tier
Start with the task you need to finish in the next ten minutes, then follow the branch. This routing is an editorial summary of the limits above, not a ranking by any vendor. Every option has a reason you might skip it.
- Do you need to avoid creating an account?
- Yes, and privacy matters most: use Duck.ai. It’s good for a quick rewrite or question with no chat storage by default. The downsides are an unpublished daily cap that resets at 1:00 a.m. WAT, and possible errors on mobile data.
- Yes, but you want a familiar chatbot: use ChatGPT signed out. You get quick answers, but no saved history and fewer features.
- No: go to question 2.
- Is the task mostly typing, such as an email, CV, caption or budget outline?
- Use ChatGPT Free. Its plain-text chat is uncapped, so long drafting sessions won’t hit a wall. The downsides are the ad programme in some markets and no access to OpenAI’s newest model.
- Choose Claude if you prefer its writing style and can batch your requests. The downside is a five-hour allowance you can use up quickly with long pastes.
- Is the task about a web page or a current fact?
- Use Perplexity’s basic search, which comes with cited sources. The downside is only 3 Pro Searches a day, and you still need to check the citations yourself.
- Is the task a document you want to upload?
- For a short file, Gemini or ChatGPT will do. Gemini’s downside is its 32k-token reading capacity, and ChatGPT’s is an unpublished file cap.
- For a very long file, split it into sections before you upload. No free plan here publishes a limit that guarantees a long document will fit.
- Is the task a translation?
- Use DeepL for a paragraph or a letter. The downside is 50,000 characters and one 5 MB file a month.
- Is the task a picture?
- For a personal image, use Copilot or Gemini. Copilot’s downsides are that it doesn’t publish its limit and that its terms say personal use only. Gemini’s downside is that images use up your chat allowance too.
- For anything you’ll sell, use ChatGPT, whose terms assign the output to you. Its free image cap is separate and not published.
Choosing the tool is only half of it. A clear first prompt uses less of every allowance above, which matters most on Claude and Gemini, where every message counts against your limit.
What to Do First
In September 2026, you can finish all five everyday tasks for free as long as you use a tool whose limit fits the task: ChatGPT for long writing, DeepL for short translations, Perplexity for cited answers, and Copilot or Gemini for personal images. Before you paste anything personal into your chosen tool, open its privacy settings and turn off model training, which OpenAI, Anthropic and Google all document. Doing this now keeps your private details out of future training data and stops you hitting a limit halfway through something urgent. Skipping it can mean an unfinished CV at midnight and your data shared more widely than you meant.
Frequently Asked Questions
What time do free AI limits reset in Nigeria?
Duck.ai resets at midnight UTC, which is 1:00 a.m. West Africa Time. Claude and Gemini refill on rolling five-hour windows, and ChatGPT’s remaining caps start counting from your first use, not from midnight.
Why does Duck.ai give me an error on mobile data?
DuckDuckGo says its anonymous way of enforcing limits can trigger errors on VPNs or cellular networks. Try switching to Wi-Fi or changing your VPN location, then report it with the Share Feedback button if it keeps happening.
Can I use a free Copilot image on my shop flyer?
Probably not. Copilot’s consumer terms say you can only use Copilot for your own personal use. ChatGPT’s terms instead assign ownership of the output to you, but check the current terms before any commercial use.
Does turning off AI training delete my old chats?
No. OpenAI says switching off “Improve the model for everyone” does not delete or hide saved chats. You have to delete chats separately if you want them removed.
