Trinka AI content detector

The free tier is credits, not words. The credits page states: "Premium users get 10 and Basic users get 4 free credits every month!", that "Unused credits lapse and are reset every 30 days", and that the AI Detector Report costs "1 credit for every 5000 words" (trinka.ai/features/credits, read 10 November 2026). The pricing FAQ adds that the Basic plan "never expires but allows only limited usage per month" and that "No credit card details are needed to register and use the Basic plan" (trinka.ai/pricing, read 10 November 2026). Four credits and 5,000 words a credit are both published; their product is not, so the card records no monthly word figure and Trinka sits in the second table of the free tiers ranking.

The pricing page prints no price in its HTML; its cards render by script from a payments API called with the visitor's country, and the figures below are that API's US dollar values for a US visitor (payments-api.trinka.ai, read 10 November 2026).

The API is priced on request: the developer portal's pricing page lists a Grammar Check tier table and says of the other APIs, "For pricing, please contact us at sales@trinka.ai", so the card records no API price per 1,000 words and Trinka is tabled apart on the API pricing ranking (developer.trinka.ai/docs/pricing, read 10 November 2026).

Retention has a number. The trust center states: "Premium user's processed data is automatically deleted from the server every 90 days. Enterprise users can opt for confidential data plan that offers real-time data deletion", and "We use AWS servers based in the US"; the pricing table repeats "Data auto-deletion 90 days" for Basic and Premium and "Real-time" for Confidential Data (trinka.ai/trust-center, read 10 November 2026). The privacy policy is open-ended: "Crimson AI reserves the right to save documents on its systems according to its policy and Indian government regulations for conducting business" and "We may delete our copies of Customer Data from time to time" (trinka.ai/privacypolicy, read 10 November 2026). The card ranks the 90 days in the data retention ranking, the reading the index gave Hive's 14-day default, and quotes the policy beside it. The trust center lists SOC Type 2, ISO 27001:2022, HIPAA, FERPA and GDPR, five standards the card carries into the compliance view, and the on-premise and offline options are the kind of deployment the offline question covers.

The claim is a leaderboard reading. Under "Why Are We #1? Proven Independently" the page prints 99.9% (accuracy) for academic abstracts, "Evaluates 12 adversarial manipulation techniques, including synonym substitution, whitespace and homoglyph injection, paraphrasing, article deletion, and case manipulation", and "With strong scores on AUROC and TPR@5% FPR, Trinka's AI detector is built for precision, not just recall" (trinka.ai/features/ai-content-detector, read 10 November 2026). The RAID leaderboard repository holds a submission named Trinka AI dated 26 May 2026 with a Crimson contact, scored on RAID's hidden test set at 0.9976 accuracy at a 5% false positive rate across all 652,800 scored generations and 1.0 on the abstracts domain with no attack (leaderboard results, read 10 November 2026). That entry is the vendor's own submission scored by the benchmark, not a paper, so it sits in this paragraph and in no rank field; the question on public leaderboard entries draws the same line.

No false positive rate is published as a figure, so Trinka has no row in the false positive ranking, and no study pairs with the 99.9%, so the claim versus evidence ranking lists the claim in its vendor table. The page's own disclaimer is the caveat the card records: the detector "is designed to assist in identifying content that may have been produced by an AI generator. However, this evaluation might not always accurately distinguish between human and AI-created content, so it should not be the sole factor in determining the nature of the content" (trinka.ai/features/ai-content-detector, read 10 November 2026). The marketplace page adds: "All AI detection tools have false positives" and "One score is a signal, not a verdict".