June 14, 2026
Copacabana Copy-Paste?
Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model
Rio’s "homemade" AI gets hit with a remix scandal as commenters split between fraud cries and "okay, but it works?"
TLDR: Rio’s new AI model was promoted as locally made, but online sleuths say it appears to be a mix of two existing public models—and its page now reportedly says it’s a merge. Commenters are split between calling that misleading, defending Brazil from cheap shots, and joking that maybe AI progress is just model remixing now.
What was pitched as Rio de Janeiro’s proud homegrown artificial intelligence chatbot brain has turned into a full-on comment-section soap opera. A GitHub issue claimed the city’s newly released Rio-3.5-Open-397B wasn’t really built from scratch at all, but looked a lot like a blend of two already-existing models—roughly 60% one model, 40% another. That alone was enough to send the internet into detective mode, with armchair sleuths, skeptics, and chaos merchants all piling in.
The loudest reaction? A mix of "this is misleading" and "wait… if the result is good, does anyone actually care?" One commenter was bluntly impressed that the mash-up seemed to perform well, basically asking whether the future of AI is just tossing open models into a blender and hoping for genius. Others were far less amused, arguing that calling it homegrown sounded suspicious if the model was mostly assembled from prior public work. Then came the backlash to the backlash: one user pushed back hard on the tone, saying accusations without proof were just bad-mouthing, and warning against lazy stereotypes about Brazilian research.
And yes, because this is the internet, the jokes got weird fast. One drive-by jab claimed Brazilians get their education and tech from Paraguay, which felt less like analysis and more like pure troll fuel. The biggest twist? A commenter noticed the model’s Hugging Face page now reportedly admits it’s a merge—a detail that only made the crowd louder, not calmer.
Key Points
- •A GitHub issue claims that Rio-3.5-Open-397B closely matches a weighted merge of existing models.
- •The alleged composition described in the issue is about 60% Nex-N2 Pro and 40% Qwen3.5-397B-A17B.
- •The issue is hosted in the nex-agi/Nex-N2 repository on GitHub.
- •The article references publicly available model materials associated with Rio-3.5-Open-397B on Hugging Face.
- •The central topic of the article is model provenance and whether the Rio model was characterized as original or derived from prior open models.