Svmuu News: The "official" open-source model Rio-3.5-Open-397B, launched by IplanRIO—an IT company under the Rio de Janeiro municipal government—has gone viral worldwide and is hailed as a rising force in Latin America’s open-source large language models. In less than 24 hours, this so-called AI “dark horse” story took a dramatic turn, as the model was accused of being a “repackaged” version created by mixing the weights of Alibaba’s Qwen 3.5 and NexN2Pro. Currently, IplanRIO has issued a public apology on the open-source community HuggingFace, admitting that it uploaded a merged baseline version for comparison "due to an operational error." The open-source model team Nex-AGI publicly accused the model from Rio de Janeiro, Brazil, of showing no evidence of having been trained by Rio itself, claiming that the so-called "in-house" model is actually a blended version of the weights from NexN2Pro and Qwen3.5-397B-A17B. (HuggingFace)