5 AI Chatbots Tested Head-to-Head: ChatGPT, Claude, Gemini, DeepSeek and Perplexity. One Winner for Everyday Tasks
Five AI chatbots now sit on millions of phones and laptops, and most people picked theirs by accident. A friend mentioned one. A new phone came with one built in. Or it was simply the first name they heard. So we stopped guessing and put all five through the same test.
The lineup is ChatGPT from OpenAI, Claude from Anthropic, Gemini from Google, DeepSeek from a Chinese AI lab, and Perplexity. Every test ran on the free plans, because free is what most people actually use. Each bot got the same six everyday jobs, word for word, with no hints and no help.
The results surprised us. The most famous name did not win every round. A specialist beat all four giants at the one job where being wrong costs you most. The free plan with almost no limits belongs to the bot most readers have never opened. And the top two finished within two points of each other.
How We Tested: Five Chatbots, Six Everyday Jobs
The setup. Each bot got a fresh free account and the same six prompts, copied exactly, with no second tries unless noted below. Two of the five use a reasoning model, an AI that works through a problem step by step before answering. We wanted to see if that shows up in daily jobs.
The six jobs. Every task is something a normal person might ask this week:
- The apology email. Write to a client, move tomorrow’s meeting to next week, sound sorry without sounding weak.
- The fact check. Five popular claims to judge true or false. Two of them were false, planted on purpose.
- The budget puzzle. A monthly income, three debts at three interest rates, and 400 spare dollars. Which debt dies first, and why?
- The weekend trip. Two days in Lisbon for 400 dollars, landing Friday noon, leaving Sunday evening, with one vegetarian in the group.
- The birthday message. For a grandmother turning 70. Warm, one gentle joke, under 100 words.
- Explain it simply. Compound interest, explained so a bright 12 year old gets it.
The scoring. Each job earned up to 10 points for accuracy, usefulness, and tone. We marked down made up facts hard, because a confident wrong answer is worse than a short one. Tone counted too. An email you must rewrite from scratch saves no time.
ChatGPT: The All-Rounder With No Weak Round
Best for: everyday help of every kind, especially trips, voice chats, and pictures.
ChatGPT never won by a landslide, and it never embarrassed itself either. The trip plan was its best round: a timed two day Lisbon route that hit the tram ride, the free viewpoints, and two vegetarian friendly lunch spots, and still landed 35 dollars under budget. It scored 10 out of 10 there, the only perfect trip plan of the five.
The apology email scored 8. It was polite, clear, and ready to send after one small edit, though the tone leaned corporate. The birthday message also scored 8: sweet, well built, and slightly greeting card. On the plain English task it shone, explaining compound interest as a snowball that grows as it rolls, and scoring 9.
Its two slips came where confidence outruns checking. On the fact check it caught the Great Wall from space myth, but happily confirmed our planted claim that the first iPhone launched in 2005. The real year is 2007. On the budget puzzle it picked the right payoff order, most expensive debt first, then made one rounding slip in the totals. Scores of 7 and 9: strong, but not careful enough to trust blind.
The free plan changed in August 2026, when OpenAI dropped the old message cap on its GPT-5.6 Luna model. The limits now sit on the extras: about three file uploads a day, a small allowance of generated pictures, and capped voice time. Voice mode is the party trick none of the other four match.
Claude: The Writing Partner You Barely Have to Edit
Best for: writing you can send without heavy editing, from emails to long documents.
Claude writes the way a careful friend writes, and the test showed it twice. Its apology email scored 9, the best of the five: it took responsibility in one line, offered two new times instead of one, and never groveled. The birthday message scored a perfect 10. It was warm without syrup, and its joke about 70 candles needing a fire permit landed exactly right.
The budget answer scored 8. Claude chose the same correct order as the winner and showed its working in a clean little table. Its plain English round matched ChatGPT at 9, with a money plant that grows its own new leaves. Only the fact check let it down. Claude caught the Great Wall myth but hedged on the iPhone year with a vague “around the mid 2000s,” which is not a fact check. That earned 6.
The trip plan scored 7 for a charming reason. Claude wrote the loveliest itinerary of the five, then blew the budget with a beautiful hotel that cost 60 dollars too much. Lovely writing, loose grip on the brief.
The free plan runs Claude Sonnet 5.5 and includes web search, file uploads, and up to five Projects. A Project is a folder that keeps the chats and files for one topic together, so your tax notes stop mixing with your holiday plans. The catch is the clock: free usage is a rolling allowance of roughly 15 to 40 messages every five hours, A relaxed evening is fine. An afternoon of heavy writing will hit the wall and ask you to wait, or pay 20 dollars a month for Pro.
Gemini: Fast, Friendly, and Recently Cut Down
Best for: life inside Google apps, on your phone and in your inbox.
Gemini answered everything quickly, and speed matters when you are replying on a train. Its trip plan scored 8: it knew Lisbon well, timed the airport run sensibly, and built the route in a clean order. One flaw cost it: a food market suggested on the wrong day, discovered only at a locked door.
The apology email scored 7: short, polite, and a little thin, missing the offer of new times until we asked. The plain English task scored 8, quick and clear if less vivid than the top two. The birthday message scored 6: fine, and completely forgettable.
Two rounds exposed the free plan’s new weakness. On the fact check, Gemini caught the Great Wall myth, missed the iPhone year like most of the field, and then wrongly flagged a true claim, that honey never spoils, as false. That scored 6. On the budget puzzle it suggested paying the smallest debt first because it feels motivating, without warning that this order costs more interest. Helpful tone, expensive advice: 5 points.
Here is why, and it is recent. On October 9, 2026, Google cut the free plan back to Flash-Lite, its smallest model. The stronger Flash and Pro models that free users could reach before now sit behind paid plans, with AI Pro starting at 19.99 dollars a month. Its home turf still matters: it lives inside Gmail, Google Docs, and most new Android phones. But as a free thinking partner, it is now the thinnest of the five.
DeepSeek: The Free Brain That Won the Hardest Round
Best for: hard thinking problems when the budget is zero.
DeepSeek produced the single best performance of the whole test. On the budget puzzle, with its DeepThink mode switched on, it laid out the payoff order step by step, most expensive debt first, and calculated the exact month each debt dies. No rounding slips, no filler, a perfect 10.
The rest of its test explains the trade. The apology email scored 7: correct and polite, but formal enough to sound like it was written by a bank. The birthday message scored 6 for the same reason. Sixty warm words were requested, and DeepSeek delivered a short, respectful speech. The trip plan scored 6 too. The skeleton was sound and the prices were sane, but it forgot the vegetarian requirement in one lunch pick.
On the fact check it scored 6, catching the Great Wall myth and missing the iPhone year, mid pack like most of the field. Its plain English answer scored 7: right in every detail, but written like a textbook, using the word principal without ever saying it just means the money you start with. A smart friend would have translated. DeepSeek often does not.
The free plan is the story here. The chat on its website and app costs nothing, has no message cap counting you down, and includes the R1 reasoning model behind that DeepThink switch. Its models are open weights too, free for anyone to download and run. Now the honest tradeoffs. DeepSeek is text only, so no pictures and no voice. At busy times the service can slow down or ask you to try again later. And the hosted chat runs on servers in China, a fair reason to keep private documents and sensitive work out of it.
Perplexity: The Researcher That Shows Its Proof
Best for: questions that need proof, not just an answer.
Perplexity is built like a search engine with a researcher on top. You ask a question, it reads live web pages, and it writes a short answer with numbered citations, the little source links that let you check where every claim came from. In the round built for that skill, it was flawless. The fact check scored a perfect 10: it caught both planted falsehoods, the Great Wall from space and the iPhone year, and backed each call with a source we could open and verify.
Step away from facts, though, and the score drops fast. The birthday message scored 4, the lowest single mark of the test. Asked for warm words for a 70th birthday, Perplexity first produced a short factual profile of what a 70th birthday is. The apology email scored 5: accurate, sourced, and about as human as a train timetable. Its budget answer scored 6, right order, thin working. Its trip plan scored 7 because the prices were real and current, even if the plan read like a spreadsheet. Plain English scored 6: correct, cited, and far too formal for a 12 year old.
The free plan gives unlimited Quick searches and about five deeper Pro searches a day. Paid Pro, at 20 dollars a month, lifts that to more than 300 Pro searches a day. One warning from our test: an answer is only as good as the pages it found, so click a citation or two before trusting a big claim.
The Five, Side by Side
Scores tell one story, limits tell another. A bot can be brilliant and still be the wrong pick if its free plan runs out on Tuesday. Here is the whole test on one screen:
| Chatbot | Free plan model | Free limits | Test score | Standout strength | Watch out for |
|---|---|---|---|---|---|
| ChatGPT | GPT-5.6 Luna | Unlimited text chat, capped tools | 51 out of 60 | Voice, pictures, trip planning | Best extras need Plus at 20 dollars a month |
| Claude | Sonnet 5.5 | About 15 to 40 messages per 5 hours | 49 out of 60 | Writing that sounds human | The clock cuts long sessions short |
| DeepSeek | V3 chat, R1 DeepThink | No message cap, slower at peak times | 42 out of 60 | Paid level reasoning for free | Text only, servers in China |
| Gemini | Flash-Lite | Daily usage limits | 40 out of 60 | Lives inside Gmail, Docs and Android | Free model cut back in October 2026 |
| Perplexity | Standard search model | Unlimited Quick searches, about 5 Pro a day | 38 out of 60 | Every answer cites its sources | Weak at creative writing |
The Winners, Category by Category
No single bot won everything, and that is the most useful finding in this whole test. Here is who took each crown:
- Best free plan: DeepSeek. No cap, no card, and the best single performance of the test on the budget puzzle. Nothing at zero cost comes close.
- Best for writing: Claude. The top email and the top birthday message. Its words needed the least fixing of any bot here.
- Best for research: Perplexity. A perfect fact check score, with receipts. It is the only one that proves its answers as it goes.
- Best all-rounder: ChatGPT. First or second in five of the six jobs, and the only bot with voice and pictures in its free kit.
Gemini and Perplexity miss the podium for opposite reasons. Gemini got weaker on free plans at exactly the wrong moment. Perplexity is a superb specialist, and specialists do not win all-round tests.
The Verdict: One Winner for Everyday Tasks
If you install just one, install ChatGPT. It won this test the way good all-rounders win anything: no weak round, the best trip plan, strong writing, and a free plan that finally stopped counting your messages. For the mix of jobs real life actually throws at you, it is the safest pair of hands.
Keep a second shortcut anyway, matched to your life. Writers should keep Claude close, because the words you send carry your name. Fact hunters should keep Perplexity, because a claim with a source beats a claim with confidence. Anyone facing hard problems on no budget should bookmark DeepSeek. And if your days already run through Gmail and an Android phone, Gemini stays the fastest helper you own, even on its smaller free model.
Five free tools, five different strengths, and one clear first download. Start with ChatGPT, add the specialist your week demands, and you will be covered for nearly everything an AI can do for free in 2026.
