ChatGPT vs Claude: We Ran the Same 3 Free Tests — Claude Never Replied Once

Search "ChatGPT vs Claude" and you'll find endless comparison tables — context windows, benchmark scores, pricing tiers. What you won't find is someone feeding the same three tasks to both free versions and showing you the answers. So we tried. ChatGPT took all three tests and answered everything. Claude never answered a single message — a brand-new free account got locked behind an ID verification wall before its first reply. Every claim below has a screenshot behind it, including the ones that didn't go our way.

Short answer: On our three identical tests — a 100-word writing task, a profit math problem, and a resume PDF summary — ChatGPT free went 3 for 3, including a self-verified math answer and a correct read of our PDF. Claude free gave us 0 answers: not because the model refused, but because our new account couldn't receive replies at all until we uploaded identity documents through Anthropic's Verification Portal. We declined — and that access wall is itself the most useful finding in this article. One honest caveat up front: verification appears to be applied selectively to some new accounts, so your experience may differ. We report exactly what happened to ours.

How we ran the test

The plan was the same one we used for our DeepSeek vs ChatGPT head-to-head:

• Three identical tasks, word for word — a product description, a math word problem with a known answer, and a document summary, copied into both tools
• Free accounts only — no Plus, no Pro, no trial, no payment details entered
• Default settings — no "Think" mode toggled on either side
• Same test file — a resume PDF we built ourselves for our Canva resume test, so we already know exactly what's in it

Then reality intervened: only one of the two tools ever received the questions. Here's the scoreboard as it actually played out.

The scoreboard

0 out of 3 for Claude isn't a judgment on its model — we never got to see it answer anything. It's a statement about what a first-day free user actually experiences, which is what most readers of a "ChatGPT vs Claude" comparison really want to know. Here's the full story in order.

Step 1: Claude's free plan looks great on paper

Registration itself was smooth. Claude's pricing page shows a genuine Free plan at $0 — and its feature list is, frankly, more impressive than ChatGPT's own free-tier description: chat across web, iOS, Android and desktop apps, code generation and data visualization, Slack and Google Workspace integration, extended thinking, and built-in web search. (One detail worth knowing: the page first showed us prices in Singapore dollars — Pro at S$30 — then switched to US dollars, Pro at $20, when we switched our connection to a US IP address. In other words, Claude's prices follow your IP region, so the currency you see may not be US dollars.)

Onboarding continued with a desktop-app upsell (skippable) and a "select your role" question (deferrable). Normal stuff. Then we opened the chat.

Step 2: The wall — "Complete verification to continue"

The first chat screen loads normally. "Welcome, P! I'm Claude." And in the bottom-right corner, the model name is right there: Sonnet 5.5 · Medium — that's what the free plan runs. So far, so good.

But the screen carries two warnings. A yellow banner across the top: "Some features are unavailable until you complete a verification for your account." And a dialog: "Complete verification to continue" — with the line that matters: "This includes replies from Claude." There's a "Not now" button. We clicked it. Nothing unlocked.

Following the "Open Verification Portal" button leads to Anthropic's Verification Portal — self-described as "Apply for and manage trust programs for your Claude usage." Not a quick email check or a phone-number text: after signing in, the portal presents a single task, "Identity verification", run by a service called Persona — the identity-checking company whose typical flow is a photo of a government ID plus a selfie.

We stopped here. Uploading a government ID to start using a free chatbot is not a decision we'll make for a tutorial — and if you feel the same, you should know the wall exists before you sign up. This is exactly the kind of detail that never survives into polished comparison tables.

Step 3: We tried to send a message anyway

Maybe the dialog was just a nag? We typed our writing prompt — the same one ChatGPT would later receive — and hit send. The message appeared in the thread with a red label underneath: "Failed to send." A notice confirmed it wasn't a network glitch:

"Your account can't receive replies from Claude right now. You can find the next step in the Anthropic Verification Portal."

Retry didn't help. The free plan's advertised features — the web search, the extended thinking, the integrations — were all behind the same door. For us, on this account, on this day, Claude free was a showroom, not a tool.

Step 4: ChatGPT free takes the same three tests

ChatGPT's free home screen shows a Free-plan account, a "Think" toggle in the input box, and — in classic ChatGPT fashion — a promo for the Codex desktop app "included for free in your ChatGPT plan." No verification anything. The three tests, one by one:

Test 1: The 100-word product description

Prompt: "Write a 100-word product description for a small handmade candle shop. Warm, friendly tone, mention free local delivery."

ChatGPT's answer: a titled piece, "Light Up Your Moments with Handmade Candles," with a body we counted at 95 words (the title adds 6 more). Slightly under the requested 100 — but it never claimed to be exactly 100, the tone is genuinely warm, and "free local delivery" is bolded right on cue. ChatGPT also auto-labeled the answer "Writing" and offered two follow-up buttons: "Make my description exactly 100 words" and "Add a stronger purchase call to action." If you want 100 words exactly, one click gets you there.

Test 2: The math problem with a known answer

Prompt: "A shop sells candles at $12 each. Materials cost $4 per candle, and the booth rental is $60 per day. How many candles must be sold to make $100 profit in a day? Show your reasoning." We knew the correct answer before asking: 20.

ChatGPT: 20 candles — correct. The reasoning is textbook-clean: Step 1, $12 − $4 = $8 profit per candle. Step 2, $100 + $60 = $160 to earn before rent. Step 3, $160 ÷ $8 = 20. And then something we like seeing: it ran a self-check at the end — revenue 20 × $12 = $240, total costs $140, profit $100 ✓. When a free chatbot checks its own work unprompted, that's a real quality signal.

Test 3: The resume PDF it had never seen

We uploaded our own test resume — the same Alex-Morgan.pdf from our Canva resume test — with the prompt: "Summarize this resume in 3 bullet points. What job is this person applying for?"

ChatGPT read the PDF's text without complaint and produced three accurate bullets: the Fashion Designer role at Borcelle from 2022 to 2024 with over 350 seasonal style guides, the key skills including apparel design and vendor negotiation with 20%+ cost reductions, and the fashion design degree with a six-month internship. To the direct question, it answered "a Fashion Designer position" and honestly noted the resume doesn't explicitly state a target job.

Two honest footnotes. First: when we built this resume we planted a "MARKETING ASSISTANT" banner near the top (it was our ATS test in the Canva article) — ChatGPT's summary didn't mention it, so a headline the human reader would spot first didn't make it into the AI's summary. "Reads the text" is not the same as "reads the document the way you do." Second: a banner at the bottom of the answer offered to unlock image analysis inside PDFs for paying users — text extraction is free; reading charts and diagrams inside files is upsold.

What we could still verify about Claude — without a single reply

Even locked out, the interface itself answered two questions people commonly have:

• What model does Claude free use? The chat screen itself displays "Sonnet 5.5 · Medium" on a free account — no downgraded legacy model name hiding in the corner.
• What's actually included free? Per Claude's own pricing page: extended thinking, built-in web search, code generation, data visualization, Slack and Google Workspace integration — across web and desktop apps. An impressive list that, for us, remained entirely theoretical.

Official claims vs what we measured

Standard caveats: one day, one account, one region. Anthropic's dialog itself says verification is requested "sometimes," so plenty of users likely never see this wall — we can only testify to the account that did. We won't extrapolate one experience into a verdict, and we'd rather publish a half-comparison with real screenshots than invent Claude's answers to fill the table.

Which one should you actually use?

What about making it a three-way test with Gemini?

"ChatGPT vs Claude vs Gemini" is the comparison people increasingly search for — and it's tempting to just add a third column. We won't, until we can hold Gemini to the same standard: same three tasks, free account, screenshots or it didn't happen. If Gemini lets a new free user talk on day one, it earns a column; if it throws its own wall, that's a finding too. Either way, we'll publish what actually happens, not a table copied from press releases.

Frequently asked questions

Is Claude free to use?

Claude does have a Free plan ($0) — and on paper it's generous: extended thinking and built-in web search are included. But in our October 2026 test, a brand-new free account could not get a single reply until we completed an ID verification through Anthropic's Verification Portal. We declined to upload documents, so we can't speak to Claude's answer quality from firsthand experience. Other users may not see this wall — it appears to be applied selectively.

Does Claude require ID verification?

In our test, yes: after registering with a free plan, Claude showed a "Complete verification to continue" dialog stating that replies are unavailable until verification is done. The portal is run by Persona and asked for identity verification — the kind that involves uploading a government ID and typically a selfie. We stopped at that step and did not submit any documents.

What model does Claude free use?

The chat interface showed "Sonnet 5.5 · Medium" in the corner of a free-plan account. Claude's own pricing page also lists extended thinking and built-in web search as included in the Free plan, alongside code generation, data visualization, and Slack and Google Workspace integration — though we couldn't confirm any of this in actual use because replies were locked.

Which is better, ChatGPT or Claude?

We can only vouch for what we measured: ChatGPT free went 3 for 3 on our identical tests — a 95-word product description close to the requested 100, a correct profit problem (answer: 20) with a self-check, and a full resume PDF summary. For Claude we obtained zero answers, so we make no quality claim either way. What we can say honestly: on a new free account, ChatGPT worked immediately and Claude didn't let us in without ID verification. That access difference is real and no comparison table mentions it.

Why did Claude say "Failed to send"?

In our case it wasn't a network problem: our message appeared in the thread with a red "Failed to send" label, and a notice explained that "Your account can't receive replies from Claude right now. You can find the next step in the Anthropic Verification Portal." Clicking "Not now" on the verification dialog did not unlock anything — replies stayed locked.

Is ChatGPT free good at math and reading documents?

In our tests, yes. It solved our candle-profit word problem correctly (20 candles) with three clearly numbered steps and an end-of-answer verification check. It also read our resume PDF's text and produced three accurate bullet points — though it identified the profile as Fashion Designer and missed a job-target banner we had planted higher on the page, and it upsells you to pay before analyzing images inside PDFs.

Wrapping up

We set out to settle "ChatGPT vs Claude" the way we settle everything: identical tasks, free accounts, screenshots at every step. The result wasn't the article we planned — it's a stranger, more useful one. ChatGPT free passed all three tests and even checked its own math. Claude free never got the chance: a new account, a verification dialog, a Persona ID wall we wouldn't cross, and a red "Failed to send" where an answer should be. If Claude's own dialog says it only asks "sometimes," plenty of readers will have a very different experience — and when we get one, we'll finish the other half of this comparison. If you want the full free-chatbot picture we've already published, read our DeepSeek vs ChatGPT six-question head-to-head, our DeepSeek beginner's guide, or see what ChatGPT's free version can do on its own. And if the resume test caught your eye, that document was built and tested in our Canva resume guide. One rule, every article: screenshots or it didn't happen.

Tested October 2026 on free web accounts of both tools, no payment details entered, no identity documents submitted. We've applied the same free-account method to free AI image generators and free AI voice tools if you're comparison-shopping beyond chatbots. Want a three-way? Our DeepSeek vs ChatGPT vs Gemini 2026 comparison runs the same-style identical tests across all three — including the round where Gemini misread our test resume.