Pangram Review (2026)The API Unit Changed Tenfold
By Daan Haket · Tested hands-on · written with an AI assistant · who does what
The unavoidable honest limit of every AI detector, including accurate ones: it is never infallible.
educators, publishers and platforms that need a low-false-positive signal on whether text is AI-generated, at web-app or API scale.
you'd use a detection score as final proof for a serious decision (no detector is safe for that), or your volume is tiny enough for the free tier.
Transparency: the “Visit” links on this page are affiliate links — we may earn a commission if you sign up, at no extra cost to you. No vendor can pay for a better review or verdict. How we make money →
A new free account, with no payment method asked for. We ran nine texts whose authorship we knew before we started: three passages from public-domain books (Darwin and William James), a paragraph one of us typed for the test, typos included, three paragraphs a language model wrote for our own reviews, one paragraph Claude wrote on the spot, and one of those machine paragraphs with three of its five sentences rewritten by Claude. We read the credit meter after every check and ran it into the daily limit.
- Nine for nine. All four human texts came back 'Human Written', including the 71-word paragraph typed for the test, and all five machine texts came back 'AI Generated', including the paragraph Claude wrote on the spot. Every result put the whole text on one side: the gauge read 100% each time. That figure is the share of the text it assigns to each side, not a confidence score.
- A machine rewrite did not shake it. With three of its five sentences rewritten by a second model, the paragraph from our Hubstaff review still scored 100% AI. That measures a model paraphrasing a model; we did not test text a person had edited by hand.
- Later the same day we gave it a tenth text: a deliberately clichéd AI paragraph after QuillBot's AI Humanizer had rewritten it (the Humanizer reported '0 AI signals'). Pangram still scored it 100% AI and added a note that the text 'appears to have been paraphrased or rewritten', pointing to one rewritten segment. Our own Claude rewrite of a review paragraph, earlier in the test, did not get that note.
Pricing (2026)
✓ Pricing re-verified 30 Sep 2026
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | up to 2,000 words a day, 3 image scans a day |
| Individual | $20/mo or $15/mo billed annually ($180 a year) | up to 300,000 words a month, 100 image scans a month |
| Professional | $65/mo or $45/mo annually ($540 a year) | up to 1,500,000 words a month, 500 image scans, $200 of API usage included |
| Team | $20 per seat/mo or $15 annually | from 2 seats, 300,000 words per seat |
| Educational Institution | no published price | — |
| Enterprise | no published price | progressive discounting beyond the developer tier |
Summarised from the pricing notes below; the notes carry the conditions.
Free: $0, no payment method, up to 2,000 words a day (20 credits of 100 words, rounded up per check: ten checks of 150 words in our test of 1 October 2026) and 3 image scans a day. Individual $20/mo or $15/mo billed annually ($180 a year) for up to 300,000 words a month and 100 image scans a month. Professional $65/mo or $45/mo annually ($540 a year) for up to 1,500,000 words a month, 500 image scans and $200 of API usage included.
Team $20 per seat/mo or $15 annually, from 2 seats, 300,000 words per seat. Educational Institution and Enterprise are both quote-only; the Enterprise card offers progressive discounting beyond the developer tier. The Developer API is prepaid pay-as-you-go: llms.txt says funds from $5 to $2,000, while the page's own top-up buttons start at $25 ($25/$50/$100/$250/$500/$1,000, with auto-refill). Realtime checks on Pangram 4 cost $0.05 per 100 words where Pangram 3 charged $0.05 per 1,000, and bulk document scans are 20% cheaper than realtime.
A 7-day free trial is offered on Individual and on Team, but not on Professional, whose button read 'Start with Professional' on 6 September and reads 'Upgrade to Professional', with '$65 billed now', on 1 October 2026. Plans change — always verify the live price on their site.
Category context: Operations & Workflow. Across the 52 tools in this category that we researched with a public US-dollar price, the median entry price is $27 a month (lowest paid plan as listed, per seat where the vendor prices per seat). Pangram’s entry price of $20 a month sits below that median. Compare pricing conditions in the Software Cost Atlas →
What the price did
We capture Pangram’s own pricing page every month and keep the figures, so this is our record of the page — not a forecast, and not someone else’s summary.
Our monthly captures caught the change that matters here. The July 2026 capture of Pangram's pricing page carries the plan prices ($20, $60, $65, $240) and nothing about per-word API rates. The August 2026 capture adds two lines that spell out the generational split: Pangram 4 at $0.05 per 100 words, and Pangram 3 at $0.05 per 1,000 words. The plan prices themselves did not move between the two, and the reading of 6 September 2026 matches August on every figure: $20/$15 on Individual, $65/$45 on Professional, $20/$15 a seat on Team, and the same two per-word API rates. The reading of 30 September 2026 matched again: $20 with $60 saved a year on Individual, $65 with $240 saved on Professional, Team at $20/$15 a seat, and $0.05 per 100 words on the API.
What the pricing page doesn’t say
It will occasionally flag genuinely human writing as AI (false positives) and miss some AI text, so it must be used as a signal, never as sole proof for high-stakes calls like academic penalties or hiring. Pangram's pitch is fewer false positives than rivals, which is real progress — but 'fewer' is not 'none'. Pair it with human judgment and the credit meter with your actual volume: the free tier's 2,000 words a day are 20 credits rounded up per check, which in our test meant ten checks of 150 words rather than twenty.
Watch the 2-minute review
The video version of this review, from our YouTube channel.
⚡ Short on time? Watch the 30-second verdict Short.
What Pangram does
AI content detector. Identifies AI-generated text with a focus on minimizing false positives — used by educators, publishers and platforms to verify whether writing is human or machine, via web app and API. The vendor's Pangram 4.0 release (July 2026) claims a false-positive rate of 1 in 10,000, detection of output from the newest frontier models, resistance to "humanized" text written to evade detection, and improved boundary detection between human and AI passages. An early research preview adds AI image detection, flagging which segments of an image appear AI-generated and verifying C2PA content credentials where present.
What buyers report about Pangram
Hands-on: tested 2026-10-01Reported by buyers: independent academic testing and user reviews, read August 2026
Themes from the buyer-review sites listed at the foot of this page, read in August 2026. We tested Pangram ourselves on 2026-10-01, but one test is a narrow sample, so here is what other buyers report. We summarise it here because the complaints are usually the part a vendor page leaves out.
- • A 1-in-10,000 false positive rate independently verified by researchers at the University of Chicago and the University of Maryland — that verification is what separates it from every detector that only cites its own numbers.
- • 99.8-100% accuracy on straightforwardly AI-generated text, with a free tier of 4 scans a day.
- • On AI-edited text — a human draft polished by a model — published accuracy drops to 73%. That is the case most real writing now falls into, and it is where accusations go wrong.
- • Users report false positives on older human-written work and on formal academic language, and reliability falls on short text (50 words minimum).
- • Public sentiment reads worse than the metrics because the people who get falsely flagged say so loudly — which is fair, since being wrongly accused costs them more than a missed detection costs anyone.
- • $20/month for 600 credits up to $65, and the credit system adds up if you also run plagiarism checks.
When the paid plan is worth it
The subscription ladder is straightforward. Free is $0 with no payment method, for up to 2,000 words a day. That is 20 credits of 100 words, rounded up per check, so at around 150 words a check it is ten checks, not twenty. Individual is $20 a month or $15 annually for up to 300,000 words a month. Professional is $65 or $45 annually for 1,500,000 words, plus 500 image scans and $200 of API usage.
Team is $20 a seat ($15 annually) from two seats. Per thousand words that is about $0.05 on Individual and $0.03 on Professional — a real volume discount. But the number that matters most is on the other side of the page. The Developer API is prepaid pay-as-you-go from $5 to $2,000, and realtime checks on Pangram 4 cost $0.05 per 100 words. That is $0.50 per thousand — ten times the Individual rate and seventeen times Professional's.
Bulk document scans are 20% cheaper than realtime, which narrows it but does not close it. So the rule is simple and it is the opposite of what most tools suggest: if the volume can go through a subscription, put it there. The API is for integration, not for saving money, and anyone routing steady volume through it is paying an order of magnitude more per word than the plan they already qualify for.
One correction that applies to older write-ups, ours included: the billing unit changed with the model generation. Pangram 3 charged $0.05 per 1,000 words; Pangram 4 charges $0.05 per 100. Any estimate built on the older figure is off by a factor of ten. And the limit no tier removes, which belongs in the buying decision rather than the footnotes: every AI detector is fallible. It will sometimes flag genuinely human writing and sometimes miss AI text.
Pangram's pitch is fewer false positives than rivals and that is real progress, but fewer is not none — so use it as a signal, never as sole proof for an academic penalty or a hiring decision.
What we found running Pangram ourselves
The first findings are at the top of this page. The rest:
- Short texts carry a caveat, and very short ones are refused. Every text of 190 words or fewer was marked 'Confidence limited — short text'; the 224-word passage was not. Below 50 words the button stays grey, with 'Minimum 50 words required for accurate detection', and nothing is charged.
- The 2,000 words a day are 20 credits of 100 words, rounded up per check. A 190-word text cost 2 credits and a 222-word text cost 3. Our first six checks came to 1,014 words and used 13 credits, the allowance for 1,300 words. At around 150 words a check, a day holds ten checks, not twenty. The cost shows next to the button before you click, and the balance sits at the top of the page.
- Hitting the limit loses nothing. With 3 credits left, a 333-word text (4 credits) could still be submitted. A window then said the entry would cost 4 credits against the 3 remaining and offered the paid plans. Nothing was scanned and nothing was charged: the balance stayed at 3. The window does not say when the credits come back, and it does not offer to trim the text to what you have left.
- Signed in and signed out, the pricing page shows the same amounts. The Professional button now reads 'Upgrade to Professional', with '$65 billed now' underneath, so that tier still has no trial.
- Not tested: text that mixes human and AI writing, text a person edited by hand, languages other than English, anything longer than 224 words, the plagiarism check (paid plans only), image detection and the API. Three of the four human texts are famous public-domain passages a detector may well have seen before. The paragraph typed for the test is the stronger evidence, and it is a single paragraph of 71 words.
Most of our reviews are researched rather than run — how we review explains which is which, and why we do not pretend otherwise.
How to actually use Pangram
Calibrate on writing you already know the provenance of, using the free tier's 2,000 words a day. Run ten pieces you wrote yourself and ten you know were machine-generated, and record what it says about each. That gives you a false-positive rate on your own material and in your own subject area, which is the only number that should govern how you use it. Then decide the policy before the subscription, because this is the honest limit of every detector including the accurate ones: it will occasionally flag genuinely human writing and it will miss some AI text.
Never use a score as sole proof for a high-stakes call — an academic penalty, a hiring decision — and write down in advance what a flag actually triggers: a conversation, not a verdict. Size the plan on words a month rather than checks: Individual covers 300,000 words for $15 a month billed annually and Professional 1,500,000 for $45, which is $5.00 and $3.00 per hundred thousand words. Team is $15 a seat annually from two seats.
If you are wiring it into something, note that the billing unit changed with the model generation. Pangram 4 charges $0.05 per 100 words in realtime where Pangram 3 charged $0.05 per 1,000 — a tenfold difference that older write-ups, ours among them, still quote the old side of. Bulk document scans run 20% cheaper than realtime, so batch whatever you can.
Where this section comes from: our research notes (165 words)
Plans, allowances and API rates read on pangram.com/pricing and pangram.com/llms.txt on 6 September 2026 from a Netherlands exit. From the page: the Individuals, Teams, Developers and Enterprise & Institutions tabs with their monthly amounts and annual savings ($60 on Individual, $240 on Professional, $60 a seat on Team), the per-card word and image-scan allowances, the Developer API panel with its top-up buttons and the two per-word rates for Pangram 4 and Pangram 3, and the trial wording on each button.
From llms.txt: the same plan list with the annual totals spelled out ($180 and $540). Unusually for this motor the machine-readable file and the live page do not contradict each other; the trial buttons and the top-up presets appear only on the page. Re-read 30 September 2026: pangram.com/pricing as served HTML (the Individuals tab: Free, Individual and Professional with their annual savings and trial buttons) and pangram.com/llms.txt for the Team, Developer API and Enterprise figures, which the served page does not carry without a browser.
The natural comparison is GPTZero or Originality.ai — close AI-detection peers — Pangram competes on lower false positives; compare accuracy claims and per-word credit cost for your volume.
Frequently asked questions
What does Pangram cost?
Free for up to 2,000 words a day. Individual is $20 a month, or $15 billed annually ($180 a year), for 300,000 words a month; Professional is $65 or $45 annually ($540) for 1.5 million words plus 500 image scans; Team is $20 per seat from two seats. Read from their own machine-readable plan list on 17 August 2026.
Is there a Pangram promo code?
We have found none, and it is one of the things people search for most about this tool. What exists instead is the annual discount, which is substantial: $15 against $20 on Individual and $45 against $65 on Professional, so a year up front saves 25% to 31%. There is also a free tier of 2,000 words a day, about ten checks of 150 words, that needs no payment method at all.
What does the API cost per document?
Watch the unit, because it changed with the model. Realtime checks on Pangram 4 are $0.05 per 100 words; the older Pangram 3 charged $0.05 per 1,000 words. Bulk document scans are 20% cheaper than realtime. Credit is prepaid, from $5 up to $2,000. A 1,000-word essay therefore costs about $0.50 on the current model, not $0.05 — check which generation any quoted rate refers to.
Is there a free trial before I pay?
Yes on two of the three paid tiers, and it is worth noting which two. Individual offers a 7-day free trial and Team a 7-day team trial, read on their pricing page on 6 September 2026. Professional, the 1,500,000-word tier, does not: its button reads 'Upgrade to Professional', with '$65 billed now' underneath (1 October 2026). So the plan whose allowance you would most want to try before committing is the one without a trial. The free plan needs no payment method and covers 2,000 words a day, which is the way to calibrate first.
Is this a hands-on review of Pangram?
Yes. We tested Pangram ourselves on 2026-10-01; the section above says what we tested, what we found and what we did not test, and where a figure comes from the vendor's own documentation instead. Most of our reviews are researched rather than run; we say which is which. We name what each tool is good and bad at, and we earn a commission only if you sign up, at no cost to you.
- Buyer reviews on independent academic testing and user reviews, read August 2026.
- Re-verified 6 September and again 30 September 2026 on pangram.com/pricing and pangram.com/llms.txt, which agree with each other on every figure.
- A dated screenshot of that pricing page, kept on this site as the receipt.
- Our own hands-on session with Pangram (2026-10-01), written up in full further down this page.
Others in this category
Reviewed operations & workflow tools that share ground with Pangram — comparable in what they do, not drop-in swaps:
New reviews, price changes we recorded, and the cost conditions we found. No sponsored placements. Unsubscribe anytime.
This is our researched assessment — not a paid placement. The “Visit” links on this page are affiliate links: we may earn a commission if you sign up, at no extra cost to you, and it never changes our take. How we review →