Pangram vs Winston AI is a choice between two paid AI text detectors that both claim near perfect accuracy, so the useful differences are price, limits and the kind of evidence behind each claim. Pick Pangram if you want a free plan that renews every day, more words per dollar, AI assisted labels and more languages. Pick Winston AI if you want AI detection, plagiarism checks and website certification in one bundle, or if a public leaderboard result under attack conditions matters to you. Winston holds a strong entry on the RAID leaderboard, where Pangram has not submitted. Pangram has more studies run by outside researchers. No independent lab has yet tested both on the same documents.

We checked the pricing pages, model and benchmark pages, RAID result files and outside studies on November 9, 2026. We did not run either detector ourselves.

Pangram vs Winston AI at a glance

ItemPangramWinston AI
Free plan2,000 words per day, renews daily, no card2,000 credits for 14 days (1 credit per word for AI checks)
Entry paid planIndividual: 20 dollars a month, or 15 dollars a month billed yearly; 300,000 words a monthEssential: 18 dollars a month, or 120 dollars a year; 100,000 credits a month
Mid planProfessional: 65 dollars a month, or 45 dollars a month yearly; 1,500,000 wordsAdvanced: 29 dollars a month, or 192 dollars a year; 200,000 credits
Top self serve planTeam: 20 dollars per seat a month, from 2 seatsElite: 49 dollars a month, or 312 dollars a year; 500,000 credits
Plagiarism checkIncluded on paid plansAdvanced and Elite only; costs 2 credits per word on top of the AI check
Current modelPangram 4.1 (October 19, 2026)Winston 5.0 (September 24, 2026)
Score styleSegments labeled Human, AI Assisted or AI GeneratedHuman Score from 0 to 100 plus a sentence prediction map
Minimum inputIntended for 50+ words; less certainty shown under 150500 characters in Winston's own evaluation
Languages24 officially supported14 listed on the homepage
Image detection3 scans a day free; 100 a month on IndividualEssential and up; 300 credits per image on the API

Sources: Pangram pricing, Pangram homepage, Pangram 4.1 model card, Winston AI pricing, Winston pricing help, Winston text analysis help and the Winston API docs.

One pricing note: on November 9, 2026, Winston's pricing page showed every paid plan at half price with the coupon code WINSTON50, labeled as a limited time promotion. The table uses the standard prices from Winston's help center. Check which one applies on the day you buy.

Free plans: daily allowance vs a 14 day trial

This is the biggest practical gap. Pangram Free gives you up to 2,000 words every day with no payment method, plus three image scans a day, file upload with OCR and the Chrome and Google Docs integrations. A teacher checking a few essays a week can stay on it. Our guide to whether Pangram Labs is free covers those limits in detail.

Winston's Free plan is a trial: 2,000 credits that last 14 days. An AI check costs 1 credit per word, so that is about two 1,000 word essays or one longer report. The trial includes the AI detector, image scans and website scans, but not the plagiarism checker, writing feedback or shareable reports. After the trial you pay or stop.

For occasional checks, Pangram wins on free use. For a wider look at no cost options, see our best free AI detector roundup.

Winston AI pricing vs Pangram pricing per word

Sticker prices look close. Words per dollar do not.

  • Pangram Individual at 20 dollars a month covers 300,000 words. That works out to about 6.7 cents per 1,000 words, with plagiarism detection available on every scan.
  • Winston Essential at 18 dollars a month covers 100,000 credits, which is 100,000 words of AI checks, or 18 cents per 1,000 words. Essential has no plagiarism checker.
  • Winston Advanced at 29 dollars a month covers 200,000 credits. If you run AI detection and plagiarism on the same text, each word costs 3 credits, so that allowance covers about 66,000 words of combined checks.

Winston's yearly prices narrow the gap (120 dollars a year is 10 dollars a month for Essential), and the current half price coupon narrows it further. Pangram's yearly Individual price is 180 dollars. Heavy scanners still get more words per dollar from Pangram. Light users who want plagiarism plus AI in one report may find Winston Advanced on a yearly plan cheaper.

For developers, Pangram's API costs 0.05 dollars per 100 words for text, with a 20 percent discount on bulk jobs, and the Individual and Professional plans include 10 and 200 dollars of monthly API credit. Winston's API uses the same credit system as the app: 1 credit per word for AI detection, 2 for plagiarism, 300 for an image, and new developer accounts get 2,000 free credits.

What each detector shows you

Pangram splits a document into segments and labels each one Human, AI Assisted or AI Generated. The AI Assisted label matters if your policy allows some AI editing but not wholesale generation. Pangram 4.1 also returns class probabilities for each segment and, by Pangram's own measurement, flips its verdict after small edits far less often than Pangram 4 did (0.15 percent of 29,103 paired edits, down from 1.08 percent).

Winston gives a Human Score from 0 to 100 percent, where a low score means the text reads as AI. Its prediction map colors each sentence in five bands from AI Generated to Human Written. Winston's own help page says the sentence map is less accurate than the overall score and tells you to use the overall score as your main signal. Winston also offers HUMN-1 certification for websites on the trial, Advanced and Elite plans, a writing feedback tool, a fact checker and an essay grader, all billed in credits.

Both offer OCR and Chrome extensions. Pangram integrates with Canvas, Brightspace, Moodle, Google Classroom and Google Docs. Winston lists Google Classroom, WordPress, Zapier and its API.

Is Winston AI accurate? What the evidence shows

Both vendors publish excellent numbers. The useful question is who ran each test and on what text.

RAID: Winston is on the board, Pangram is not

RAID is the public benchmark for AI text detectors from the University of Pennsylvania, explained in our RAID benchmark guide. Winston's entry covers the model it released on September 24, 2026. We read the results file in the RAID repository: with all adversarial attacks included and the false positive rate held at 1 percent, Winston caught 652,073 of 652,800 AI texts, or 99.89 percent, the highest of the 57 entries we checked on November 9, 2026. With no attacks it scored 99.998 percent. Pangram had no submission.

Two caveats. RAID's AI text comes from models up to GPT-4, not 2026 chatbots. And RAID sets its own cutoff per writing domain to hold false positives at 1 percent. For book passages the cutoff it chose for Winston was about 0.84, well above the 0.5 cutoff Winston says its app uses, which implies the default app setting would flag more than 1 percent of RAID's human book passages. The leaderboard number is a fair comparison between detectors, not a promise about the app's default verdicts.

The Chicago Booth benchmark: same data, different runs

Brian Jabarian and Alex Imas at the University of Chicago Booth School of Business built a corpus of 1,992 human passages across six genres, each matched with versions from four frontier models. Their NBER working paper tested Pangram, Originality.ai, GPTZero and an open source RoBERTa model. It did not test Winston. It found that Pangram was the only detector to meet a strict false positive cap of 0.005 without giving up accuracy, including on very short passages and on text passed through the StealthGPT humanizer.

Since then the vendors have rerun the data themselves:

  • GPTZero's January 2026 rerun, as quoted on Winston's benchmark comparison page, reported Pangram at 98.9 percent AI recall with one false positive out of 1,992 human texts.
  • Winston's own September 2026 run of Winston 5.0 reported 100 percent on all 8,697 documents it scored, per its Winston 5.0 report. Winston removed texts under its 500 character minimum first, so it scored 1,747 human and 6,950 AI texts rather than the full 1,992 and 7,968. The paper found that errors rise on short passages, so those dropped texts were among the harder cases.
  • Pangram's 4.1 model card reports 0 percent false positives and 0 percent false negatives on 15,936 full length documents from the same benchmark.

So both current models claim a perfect score on this benchmark, both from runs the vendors did themselves. Only Pangram has an independent result here, and that one was for an earlier model.

Studies that included Pangram but not Winston

Pangram appears in more outside studies. In Epoch AI's July 2026 test, Pangram flagged none of 495 human passages and caught every plainly prompted AI passage, but missed 30 of 297 passages where the model imitated a real author's style. A 2026 study from Vrije Universiteit Brussel of 160 academic papers (VUB research portal) found Pangram ahead of GPTZero, Copyleaks and Turnitin on fully AI, hybrid and humanized papers. Winston was not in either test.

Student essays and non native writers

Both vendors report zero false positives on all 25,996 PERSUADE 2.0 student essays (Winston's own run, and Pangram's as Winston reports it). Winston adds that the result held across every grade, gender, ethnicity, income and disability group in that corpus. Pangram's model card reports 0 false positives on the 91 non native English essays from the 2023 Liang et al. study that once caught out many detectors. Both are vendor measurements.

The honest summary: Winston has the stronger public leaderboard result and very strong self reported numbers on current models. Pangram has the longer record with researchers who do not work for it. Our are AI detectors accurate guide explains why neither kind of number should decide a misconduct case on its own.

Who should pick Pangram

  • Teachers and editors who check a few documents a week. The daily free allowance covers it.
  • Schools on Canvas, Brightspace or Moodle. Pangram's institutional plan plugs into all three.
  • Anyone who needs mixed authorship labels. AI Assisted segments fit policies that allow light AI editing.
  • Multilingual teams. Pangram lists 24 supported languages, including Arabic, Japanese, Korean, Hindi and Urdu, against Winston's 14.
  • High volume scanning. About 6.7 cents per 1,000 words on Individual, with plagiarism included.

If you are weighing Pangram against the other big consumer detector, see Pangram vs GPTZero.

Who should pick Winston AI

  • SEO teams and publishers on WordPress. The WordPress and Zapier integrations and HUMN-1 website certification suit content teams.
  • Buyers who want AI and plagiarism checks in one report and scan modest volumes, on Advanced or Elite.
  • People who weigh public leaderboards heavily. Winston had the top RAID score with attacks on as of November 9, 2026.
  • Short trial users. If you only need a few checks this fortnight, the trial and the current coupon make Winston cheap to try.

How to choose between them in an afternoon

  1. Collect 20 or more texts you know were written by people in your setting, including short ones and any from second language writers.
  2. Add 20 AI texts from the models your writers actually use, some lightly edited.
  3. Run them through Pangram Free and Winston's trial. Note every human text flagged and every AI text missed.
  4. Compare how each tool shows mixed documents.
  5. Price your real monthly volume on both pricing pages, including plagiarism if you need it.

Treat any flag as a reason for a conversation, not as proof.

FAQ

Is Pangram better than Winston AI?

It depends on what you value. Pangram has a renewing free plan, more words per dollar, AI Assisted labels, more languages and more independent studies. Winston has a lower yearly entry price, bundled plagiarism on its upper plans and the top RAID score. Neither has been tested against the other on the same documents by an outside lab.

Is Winston AI accurate?

On public benchmarks it scores very well. Its RAID entry caught 99.89 percent of attacked AI text at a 1 percent false positive rate, and Winston reports perfect results on the Chicago Booth and PERSUADE 2.0 data. Those last two are Winston's own runs, and RAID uses older AI models, so test it on your own writers before you rely on it.

Is Winston AI free?

Only as a trial. The Free plan gives 2,000 credits for 14 days, and an AI check costs 1 credit per word. Pangram's free plan renews at 2,000 words every day.

How much does Winston AI cost?

Standard prices are 18, 29 and 49 dollars a month for Essential, Advanced and Elite, or 120, 192 and 312 dollars a year. On November 9, 2026, the pricing page showed a limited time half price coupon.

Does Pangram or Winston AI check for plagiarism?

Both do. Pangram includes plagiarism detection on its paid plans. Winston includes it on Advanced and Elite, at 2 credits per word on top of the AI check.

Which is better for non native English writers?

Pangram's model card reports no false positives on the 91 non native essays from the Liang et al. study, and Winston reports no false positives across all PERSUADE 2.0 student groups. Both are vendor tests. If your writers include many second language students, run your own sample through both before you choose.