Pangram 4 claims 1 error per 24k docs — but detection is a cat-and-mouse game
pangram released pangram 4, an ai text detector claiming a 99.66% detection rate and one false positive per 24,000 documents, while raising api prices. this matters as the industry struggles to distinguish human writing from ai-generated content reliably.
pangram introduced pangram 4, a new ai detection model that claims to identify 99.66% of ai-generated text with a false positive rate of 0.0041%, or roughly one error per 24,000 documents. the company states the model distinguishes between pure ai output and human writing mixed with ai assistance, detecting the ai component in 98.83% of texts modified by common 'humanizer' tools. these performance gains come with a significant price hike, bringing the api cost to $0.05 per 100 words, while legacy support for pangram 3 ends in september 2026. this launch signals continued demand for verification tools as institutions seek to manage ai integration without relying on unreliable blocking methods.