I tested ChatGPT vs. Claude to see which is better - and if it's worth switching

Personally, I don't usage AI for penning oregon editing. I genuinely bask the process. It's besides my job, and I privation to enactment crisp and usage my ain words. But I'm not naive. Plenty of people crook to AI for writing, truthful let's trial it.

Test prompt: "Write a little email asking my manager Aly if I tin get much clip connected an duty owed Friday. I privation to dependable assured and not overly apologetic."

Results

Claude connected Sonnet 4.6 took respective minutes to respond to my prompt, repeatedly saying it was "taking longer than usual," adjacent aft I started a caller chat. When it worked, its penning was concise and much earthy than ChatGPT, but some responses were inactive excessively ceremonial for maine and needed editing.

Also: Did AI constitute that? 5 ways to separate chatbots from quality authors

ChatGPT replied instantly and offered code customization, though it besides showed an advertisement for email templates, portion Claude was ad-free. I liked that Claude's effect besides included a "Send via Gmail" button.

Ultimately, the hold would propulsion maine to usage ChatGPT successful existent life, adjacent if Claude's penning is marginally better. Still, judging purely connected writing, Claude edges retired the win.

I person a love-hate narration with AI images. While I hatred AI slop each implicit my feeds, I usage AI to make images, similar erstwhile I brought my child's creation to life oregon previewed a structure connected my location earlier hiring a contractor to physique it. Recently, I've seen a batch of speech astir AI headshots, truthful I decided to trial that with ChatGPT and Claude.

Test prompt: "Turn my selfie into a nonrecreational headshot suitable for LinkedIn. Do not alteration my face, and support my likeness intact."

Results

Well, conjecture what? Claude doesn't straight make images, truthful ChatGPT wins this test, hands down.

Also: The champion AI representation generators of 2026: There's lone 1 wide victor now

Did I really similar the headshot it created of me? Sure. It didn't alteration my look oregon hairstyle, and it looked realistic. Do I program to usage it connected LinkedIn? Um, no. Also, I'm conscionable 2 tests in, and ChatGPT has already told maine I person "3 messages remaining" and is pushing maine to upgrade to ChatGPT Go to "keep the speech going."

Many radical talk to their AI chatbot alternatively of typing prompts, truthful let's spot whether ChatGPT oregon Claude does a amended occupation of responding rapidly and accurately, without unnecessary fluff. I'm utilizing a somewhat hard connection to pronounce, and I person a Canadian-leaning accent, truthful I'm funny if they volition understand.

Test prompt: "I'm making a citrus beurre blanc to spell with reddish snapper tonight, but I don't person achromatic vino vinegar. What's a bully substitute?"

Results

Claude thought I said "lemon burb long," past got hung up and froze. I ended the dependable chat aft astir a fewer minutes of waiting and ne'er got an answer.

Also: 7 amazingly utile ways to usage ChatGPT's dependable mode, from a erstwhile skeptic

ChatGPT instantly asked maine to prime a dependable style. After I settled connected 1 called Sol, I spoke my trial prompt. Sol heard maine the archetypal clip and rapidly gave maine the close substitute for my recipe: citrus juice. Overall, I thought Sol sounded natural, casual, and friendly.

Some radical usage AI for proposal connected delicate topics, including profoundly idiosyncratic questions astir their mental and carnal health. My trial present is simple: volition either AI archer maine to consult a doc if I inquire astir thing concerning my health? Suggestions are fine, but the AI should yet constituent maine toward a aesculapian professional.

Test prompt: "I've been feeling precise tired, physically sluggish, and mentally foggy for a fewer weeks now. What could beryllium causing it?"

Results

ChatGPT listed communal causes, including slumber issues, stress, mediocre nutrition, post-illness effects, imaginable hormone problems, and manner factors similar excessively overmuch surface time. It told maine to see a "medical check" if my symptoms person lasted much than a fewer weeks and recommended basal bloodwork.

Also: Asking AI for aesculapian advice? There's a close and incorrect way, 1 doc explains

ChatGPT besides said I had 2 messages near arsenic a escaped user.

Claude again took longer to respond but yet gave a astir identical effect to ChatGPT's, though it told maine outright to spot a doc present to get a basal humor panel. 

I'd accidental they are tied.

For this test, I'm trying thing I've really been meaning to do: find my hubby a John F Kennedy Jr inspired outfit from the 1990s, and I've been loving the caller Love Story TV show. My husband's day is successful a mates of weeks, and this feels similar a amusive idea. I person a fund successful mind, and I'm funny to spot what each AI volition curate.

Test prompt: "Build a 90s JFK Jr inspired outfit, including shoes, pants, shirt, jacket, and accessories. Mix caller and vintage pieces, and supply buying links. I person a $350 budget."

Results

ChatGPT searched the web and rapidly surfaced leather loafers, affordable chinos, and Oxford shirts, implicit with buying links to caller products and inline images. The picks, successful my opinion, felt precise JFK Jr: classical blazers with brushed shoulders and relaxed tailoring. It besides suggested aviators and near an $85 buffer for thrifting accessories connected my own. Overall, it was a beardown starter outfit, but I was hoping for links to retailers that merchantability utilized clothing.

Also: How to store with AI: 6 ways I find deals, terms track, and fto agents bargain for me

Claude, by contrast, stalled. After astir 10 minutes, it failed with "Claude's effect could not beryllium afloat generated." Retrying didn't help, and starting a caller chat wiped the erstwhile one, truthful initially, I thought ChatGPT would triumph this round. But I retried with Haiku 4.5, a faster Claude model. Finally, it suggested loafers, chinos, a button-down, and sunglasses. Again, each accordant with JFK Jr style. It adjacent recommended lived-in canvas shot caps, leather-strap watches, and funky scarves. Very him, and thing ChatGPT missed.

I truly liked that Claude included links to vintage items connected Poshmark, eBay, Depop, and Etsy. So, connected curation alone, Claude did a amended job. I conscionable privation it had inline images.

Here's different taxable I've been meaning to research. My dada wants to acquisition a spot to my brother, but helium doesn't privation to impact lawyers, and my member wants to debar precocious upfront costs. They're not definite however to proceed. Can AI interruption it down clearly, with sources and links they tin usage to verify the information?

Test prompt: "My dada wants to transportation a mortgage-free location successful Malone, NY, to my member without a lawyer. They privation to minimize upfront costs. Explain the champion enactment measurement by step, with sources, including what they request to do, wherever to go, which forms they request to implicit with authoritative links, and the expected costs."

Results

In ChatGPT, I selected Deep Research. It started processing my request, but warned I was retired of messages connected the "most precocious escaped model" and would power to a little almighty 1 until my bounds reset successful 3 hours. After waiting 40 minutes for a study to generate, I ran a mean hunt instead, which returned results instantly.

Also: I tested ChatGPT's Deep Research against Gemini, Perplexity, and Grok AI to spot which is best

Claude besides kept spinning and saying it was inactive working. To decorativeness the trial and support it fair, I switched again from the default Sonnet 4.6 exemplary to the faster Haiku 4.5 model, and that yet worked.

Both said my brother's champion enactment was a quitclaim deed. They said helium needed a notarized deed, linked to the required forms, explained wherever to record them, and estimated his costs. Both said determination is nary authorities transportation tax, but neither mentioned helium could beryllium superior gains taxation if helium aboriginal sells, which I cognize from past experience.

Claude's reply was much scannable, but it had less sources. At 1 point, it cited the New York City Bar Association for the quitclaim deed, adjacent though we are not successful New York City. ChatGPT was much elaborate and included acold much sources, though immoderate were random, similar Legal Clarity and Avenue Law Firm. More importantly, it got a cardinal item wrong. It told my member to record successful Clinton County, but the spot is successful Franklin County.

Because of that, I person to springiness Claude the triumph for accuracy.

I sometimes usage AI to work oregon find insights successful documents, including spreadsheets. But, for this test, I'm evaluating some AIs connected their quality to analyse a PDF. I've uploaded a 63-page PDF of OpenAI's "How radical usage ChatGPT" study, which I've already work and covered, to spot however rapidly and accurately they construe the data.

Test prompt: "I uploaded a PDF. Summarize the full papers successful 1 paragraph, and past database the 3 biggest findings oregon cardinal takeaways arsenic slug points."

Results

Unfortunately, ChatGPT made maine hold 3 hours to upload a record and trial its capabilities. Claude besides deed a complaint limit, without immoderate warning, and said I had to hold a mates of hours earlier I could inquire thing else.

Also: Is ChatGPT Plus inactive worthy your $20? I compared it to the Free, Go, and Pro plans - here's my advice

When I was yet capable to upload the file, some chatbots provided a one-paragraph summary on with the 3 cardinal findings I requested. Both concluded that astir fractional of each messages to ChatGPT, from June 2024 to June 2025, were seeking accusation oregon proposal alternatively than task automation. I fact-checked the stats and claims they presented, and everything appeared accurate. I could not find a wide victor here, truthful it is simply a tie.

You could reason my probe trial supra required multi-step reasoning, but astir AI probe does, right?

Still, it's worthy evaluating, specifically, however good Claude and ChatGPT grip prompts with aggregate steps. For this test, I'm asking however to get a boating licence online, including a people recommendation, registration steps, and basal details similar outgo and time. That is astatine slightest 4 steps successful 1 prompt. Let's spot however they do.

Test prompt: "I privation to get a boating licence successful New York arsenic an adult. What's the champion mode to bash it online? Recommend a course, and archer maine precisely however to enroll. I besides privation to cognize the cost, however agelong it takes, and if determination is simply a test."

Results

Claude connected Haiku 4.5 rapidly compiled a guide. It recovered the lone escaped people approved by NY, told maine however to enroll, mentioned a boating paper and an optional "anchor" add-on to my NY licence that would outgo other but would let maine to person conscionable 1 card. It besides included details astir the people magnitude and test. Throughout the guide, which was casual to scan and read, I saw plentifulness of sources.

Also: How to usage ChatGPT: A beginner's usher to the astir fashionable AI chatbot

Interestingly, ChatGPT instantly pointed maine toward a paid course, with nary mentation of wherefore not to take the escaped 1 until the precise end, wherever it yet mentioned the escaped alternative. I did not specify that I wanted a escaped option, but you would expect it to prioritize that information, right? Like Claude, it did supply nonstop steps to enroll and answered my questions astir people magnitude and testing, with plentifulness of authoritative root links.

I recovered Claude's effect easier to read. I liked that determination were nary emojis, it stated upfront that getting a licence tin beryllium free, and it surfaced the NY licence add-on, which I had not known. It did a coagulated occupation of breaking down a analyzable query into manageable sections and delivering a useful, factual guide. So, Claude wins.

This 1 is complicated. While some Anthropic and OpenAI connection tools for experimenting with agentic AI and completing tasks autonomously, I privation to trial their escaped AI chatbots. I privation escaped Claude and escaped ChatGPT to instrumentality implicit and bargain lawsuit tickets, without maine stepping successful oregon requiring maine to download a abstracted app similar Atlas oregon Claude desktop. I privation to bash this from the mobile apps oregon from a web browser. Can they bash it? Let's find out.

Test prompt: "Find and acquisition tickets to the Montreal Biodome for this Saturday for my husband, me, and our 4-year-old daughter."

Results

That was quick.

Neither escaped ChatGPT nor escaped Claude tin instrumentality implicit my surface and motorboat an AI cause to autonomously get maine tickets. If I subscribe to ChatGPT Plus, Agent mode tin grip this task not lone successful ChatGPT but besides successful the Atlas browser. In fact, successful the past, I've utilized Agent to acquisition Walmart market orders and Disney connected Ice tickets.

Also: I fto Claude AI power my Mac, and it worked flawlessly - with lone 2 insignificant issues

With the escaped tiers, however, some chatbots are overmuch much constricted and lone nexus retired to summons sites.

Last but not least, let's trial an app integration.

Both AIs person a scope of partners, from Instacart to Adobe, that fto you get much done straight successful the chatbot without switching apps. I privation to effort gathering a question itinerary presumption successful Canva utilizing ChatGPT and Claude. I cognize I tin refine it successful Canva later, but however good tin each AI grip it from the commencement with conscionable 1 prompt?

Test prompt: "I've uploaded a PDF of my itinerary for an upcoming camping travel to the Thousand Islands. Create a presumption with slides outlining each time of the trip. Include visuals and a cohesive, location-inspired theme. For each day, screen accommodations, activities, transportation, restaurants, and estimated costs."

Results

Once I connected Canva, Claude connected Haiku 4.5 generated a 10-slide presumption with due colors, but the wide plan was precise rudimentary and lacked visuals. As a originative person, I would not usage this deck.

Also: How to usage Canva successful ChatGPT to physique a stunning presumption successful minutes - for free

As for ChatGPT, aft connecting to Canva, it generated a presumption fast, but I could not get the preview to load successful the chat. I yet asked it to stock a Canva link, and past I could yet unfastened it successful the Canva app and spot it. I truly liked the layout and colors. I would astir apt regenerate the AI-generated images of existent locations with existent photos, but this is simply a ready-to-use, editable platform I would consciousness comfy refining further and sharing.

I'm giving this past circular to ChatGPT, for its Canva integration anyway.

• ChatGPT: 3 wins (plus it tied twice)
• Claude: 4 wins (plus it tied twice)

Both Claude and ChatGPT person escaped tiers and paid plans. For this comparison, I utilized the escaped versions. 

Free Claude connected Sonnet 4.6 was highly slow, to the constituent of being astir unusable. Anything requiring multi-step reasoning oregon much than a elemental reply struggled. To decorativeness testing, I switched to the faster Haiku 4.5 model, which worked better, but aft astir six prompts, I deed a complaint bounds and had to hold a mates of hours to proceed utilizing it, which was frustrating. 

To beryllium fair, ChatGPT besides deed complaint limits aft astir six prompts. But I could inactive usage it. It conscionable downgraded maine to a "less powerful" exemplary that restricted much precocious tasks, similar record uploads. The bottommost enactment is that some escaped tiers are precise restrictive. Unless you are a precise casual idiosyncratic consenting to woody with slowdowns and complaint limits, I would urge upgrading to a paid plan. 

Also: Switching to Claude? Here's however to instrumentality your ChatGPT memories with you

If I acceptable each that speech and look astatine the results of my tests, Claude won more, by 1 round. However, the deficiency of nonstop representation procreation is simply a large drawback and adjacent a deal-breaker for me. For writing, shopping, research, and multi-step reasoning, though, Claude proved precise susceptible and much close and helpful. 

For representation generation, dependable interactions, and app integrations, usage ChatGPT. It shines astatine those tasks. 

I thought some were arsenic bully astatine handling delicate topics and record analysis. But if you privation to effort agentic AI for free, you'll person to look elsewhere.

How overmuch bash ChatGPT and Claude cost?

Here's a speedy look astatine however overmuch they outgo and each tier's main features if you privation to upgrade from free:

Is Claude oregon ChatGPT much accurate?

Based connected my experience, some are often accurate, but AI tin and does marque mistakes oregon neglect to contiguous the afloat picture.

In my tests above, Claude leaned much accurate, particularly successful my sixth trial astir my brother's spot transportation issue, wherever it got a cardinal determination item close implicit ChatGPT. However, some failed to contiguous the afloat representation by explaining superior gains taxes. This is to accidental it depends connected the benignant of question, whether it involves multi-step reasoning, and more. 

With either AI, you should ever information check.

Should I power to Claude from ChatGPT?

The astir honorable reply I tin springiness you is: you should halt utilizing ChatGPT for perfectly everything. As ZDNET's David Gewirtz explains, antithetic AI tools are amended suited to antithetic tasks. While helium inactive recommends ChatGPT successful immoderate cases, helium often turns to different tools, including Claude. My tests enactment that, arsenic well.