(Hmm, the familiar "temporary worker" feeling😅😅)

Latin America's "self-developed dark horse" turned into a shell joke overnight, Alibaba Qianwen becomes the biggest winner? 🇧🇷🇨🇳
The "official" open-source model Rio-3.5-Open-397B launched by IplanRIO, an IT company under the Rio de Janeiro city government, has gone viral worldwide and is regarded as a new force in Latin America's open-source large models. In less than 24 hours, this so-called AI "dark horse myth" was reversed, with the model accused of being a "shell" version that mixes weights from Alibaba's Qwen3.5 and NexN2Pro.
Open-source model team Nex-AGI was the first to come forward publicly to debunk: in this so-called self-developed model, there is no trace of Rio's official training at all. After technical disassembly, it turns out to be a "patchwork" created by merging weights from the open-source NexN2Pro model and Alibaba's Qwen3.5-397B-A17B!
💬Rio's official company IplanRIO has urgently issued an apology on the open-source community HuggingFace. However, their public relations statement is very subtle, claiming it was "due to operational error" and that they accidentally sent the wrong merged baseline version used for comparison testing...
#AI #Large Model #开源 #Qwen #阿里千问 #HuggingFace