AI detectors ranked by the generator models the vendor names

Field sorted: the generator count field on each card, written on 9 November 2026 from the generators list that was read on 8 November 2026 for the first sixteen cards and 9 November 2026 for the twenty-six added that day, each from the vendor page recorded on the card. A name counts once when the vendor lists it as a model, a model family or a product whose output the detector scores or was trained against: ChatGPT, GPT-4 and GPT-5 are three names because the vendor lists three; "DALL-E 2 and 3" is one name because the vendor lists one item with two versions; Turnitin's 35 are the 33 rows of its version table plus the GPT-3.5 and GPT-4 generators its Spanish and Japanese models name. A company name counts once only where no model of that company is named on the same page (Proofademic's research page). Catch-all phrases ("other LLMs", "many others", "and more"), headline counts ("35+ LLM families", "500+ generators", "26 models benchmarked", "about 90 further classes") and announced partners ("coming soon") are printed in the row and not counted. Direction: more names rank higher. Tie rule: more study records first, then product name, so Hive (2 records) sits above Sightengine (1) at 33, Copyleaks (9) above AI or Not (1) at 15, ZeroGPT (7) above Sapling (3) above Phrasly (0) at 11, GPTZero (16) above Originality.ai (7) above Winston AI (2) at 10. 56 of 72 qualify; 16 name no generator and are tabled apart. Full rules: methodology page. The count says how many names a vendor prints, never how many models a detector catches; the independently tested ranking and the GPT-4 question say which generators a study actually ran.

A vendor can print a headline count, a named list, or both, and the two rarely agree. ZeroGPT names eleven models and states "35+ LLM families" (zerogpt.com, read 8 November 2026); TruthScan names thirteen and states "500+ generators" for images (truthscan.com, read 9 November 2026); Pangram names sixteen and states "26 models benchmarked" (pangram.com, read 8 November 2026); Resemble Detect names 62 and states "250+" (resemble.ai, read 9 November 2026). This page counts the names, because a name can be checked against a model that exists and a headline count cannot. Turnitin lists versions where other vendors list families, so its 35 rows cover fewer families than GPTZero's ten names (guides.turnitin.com, read 8 November 2026); the page counts what each vendor chose to print. Hive publishes class identifiers (sora2, gptimage1_5) in its documentation rather than marketing names, and the count follows the documentation (docs.thehive.ai, read 8 November 2026). Writer and BrandWell are counted from archived pages because the live pages redirect, read 9 November 2026. All read 8 and 9 November 2026; the Updated line is the date the page last changed.