Pangram Labs
AI-text detector with the lowest false-positive rate in a Chicago study
Overview
Pangram, from Brooklyn-based Pangram Labs, detects AI-generated and AI-assisted text segment by segment and, since July 2026, AI-generated images. It is sold as a free plan, $20 and $65 monthly subscriptions, a pay-as-you-go API and institutional licences with LMS integrations. A University of Chicago working paper found it had the lowest false positive rate of the detectors it tested.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- AI-assistance detection
- 3 image scans a day
- 300,000 words a month
- 100 image detection scans a month
- Similarity detection on every scan
- Social feed scanning
- File upload and OCR
- Google Docs and Gmail integration
- Everything in Individual
- Admin and member dashboard
- Unified billing
- Everything in Individual
- 5x more words
- 5x more image scans
- $200 of API credits a month
- Unlimited AI and similarity checks in the LMS
- Canvas, Brightspace, Moodle and Google Classroom
- Usage statistics
- No training on student data
- API access
- Progressive volume discounts
- SOC 2 compliance
- Shared development and support channel
Use cases
What you can produce with Pangram Labs
- Segment-by-segment labels marking passages as human, AI-assisted or AI-generated, with a document summary of Human, Mixed or AI
- A humanizer score flagging text run through AI paraphrasing tools
- AI-generated image detection with a heat map over the suspect region
- Similarity checking against copied and re-used text on paid plans
- API JSON with per-segment offsets and confidence, at $0.05 per 100 words
- In-LMS checks for Canvas, Moodle, Brightspace and Google Classroom on the institutional licence
- A browser extension that labels posts in social feeds
ASEAN Perspective
Pangram Labs in Southeast Asia
Pangram's multilingual support covers 20+ languages but is anchored in English accuracy; detection quality for ASEAN languages such as Bahasa Indonesia, Tagalog, Thai, and Vietnamese has not been independently validated, making it a less certain choice for regional content-moderation workflows. There is no Asia-Pacific office, local data residency option, or regional pricing, which may create friction for Singapore and PDPA-compliant enterprise buyers. Its SOC 2 Type II certification and ephemeral data processing align broadly with PDPA and ASEAN data protection principles, but buyers should confirm no content is routed through US-only infrastructure before deployment. For ASEAN publishers and trust-and-safety teams primarily working in English — such as regional news aggregators and English-language academic institutions — Pangram's best-in-class false positive rate and API reliability make it a credible option worth evaluating.
Pangram has the strongest outside evidence of any commercial AI detector we list: a University of Chicago working paper found its false positive rate essentially zero on its test corpus, lower than GPTZero and Originality.ai, though all three stayed under 1%. Its own model card claims about one false positive in 24,000 English documents for Pangram 4. Segment-level labels separating human, AI-assisted and AI-generated passages are more useful than a single percentage.
Treat a flag as evidence, not proof: a TechCrunch test saw it mislabel rewritten and human sentences. Pricing is fair, the free plan is generous, and the API is cheap for moderation pipelines.
What we found
Pangram is an AI-text detector from Pangram Labs, Inc., a Brooklyn company set up in 2023 by Max Spero and Bradley Emi, who met as Stanford undergraduates. Spero previously worked at Nuro, Google, Two Sigma and Yelp; Emi led deep-learning research at Absci and worked on Tesla Autopilot. The company traded as Checkfor.ai until April 2024.
The accuracy story has two layers, and they should be kept apart. Pangram's own model card for Pangram 4, released on 29 July 2026, reports a 0.0041% false positive rate on a million human-written English web documents, about one in 24,000, and a 50-word minimum input; the homepage still quotes one in 10,000. The outside evidence is a 2025 University of Chicago working paper by Brian Jabarian and Alex Imas, which tested Pangram against GPTZero, Originality.ai and an open-source RoBERTa model. Pangram had the lowest false positive rate, essentially zero across most thresholds, although all three commercial tools stayed under 1%. That paper has not been peer reviewed. The EditLens method behind Pangram's mixed-authorship scoring was accepted at ICLR 2026.
No detector is infallible, and a flag should open a conversation rather than settle one. Testing it for TechCrunch in July 2026, reporter Rebecca Bellan found Pangram caught fully AI-written articles easily, but it also flagged sentences the reporter had rewritten by hand as AI and marked some human-written sentences as AI-assisted.
The free plan gives 20 checks a day after sign-up. Individual costs $20 a month, or $180 a year, for 300,000 words, 100 image scans, similarity checking and Google Docs and Gmail add-ons; Professional at $65 a month raises the limits fivefold and includes $200 of API credit. The API charges $0.05 per 100 words. Schools buy an institutional licence with Canvas, Moodle, Brightspace and Google Classroom integrations. Pangram raised $9 million led by Menlo Ventures in July 2026 and launched an image detector alongside Pangram 4.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- Pangram's founders — Max Spero and Bradley Emi — met as Stanford freshmen and both went on to work in AI at Google and Tesla before co-founding the company.
- EditLens, the model powering Pangram 3.0, was accepted at ICLR 2026 — making Pangram the first commercial AI-detection company to publish at a top ML conference.
- When Pangram scanned ICLR 2026 peer reviews for the conference itself, it predicted that approximately 21% of submitted reviews were AI-generated — a finding covered by Nature News and The Atlantic.
- Quora used Pangram's API to scan over one million posts, reporting zero false-positive complaints in production — a real-world stress test that independent benchmarks alone cannot replicate.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Pangram Labs's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Pangram Labs unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Pangram Labs directly →
Spotted something out of date? Suggest an update →
Alternatives to Pangram Labs
More in Security & Safety
