Show HN: Researchers say they fingerprinted 178 AI models and found clone clusters, cross‑provider twins, and huge price gaps

What they did

It has been reported that a team behind Rival.tips ran a large‑scale stylistic audit of 178 generative AI models, measuring outputs across 32 writing dimensions. The study allegedly analyzed 3,095 model responses to build fingerprints for each model — think of it as biometric data for prose. Short prompts, long prompts, and everything in between were compared to tease out stylistic signatures.

What they found

It has been reported that the analysis uncovered "clone clusters" and surprising cross‑provider twins — distinct models that nevertheless write almost indistinguishably. Even sharper: some models that produce nearly identical text reportedly differ in price by as much as 185x. That gap raises an eyebrow. Who's paying for branding, and who’s paying for real utility?

Why it matters

If true, the findings speak to two fast‑moving trends in AI: commoditization of core capabilities, and a growing need for transparent benchmarks. How should buyers choose between a cheaper twin and a premium offering that sounds the same? The work, shared on Hacker News and linked from Rival.tips, is a reminder that as models proliferate, independent audits and stylistic forensics will become key tools for procurement, compliance, and research.

Sources: rival.tips, Hacker News