Scandal around Rio 3.5: "Municipal" AI model turns out to be a clone of Nex N2 Pro
The municipal IT company of Rio de Janeiro, IplanRIO, presented the open AI model Rio 3.5 Open 397B, claiming it surpasses DeepSeek V4 Pro and Qwen 3.7 Plus. However, just a day after the announcement, the development team Nex published evidence that the "new" model is a direct merge of their own development, Nex-N2-Pro, and Qwen3.5-397B-A17B. This story is a vivid example of how loud claims about technological sovereignty shatter against the harsh reality of open-source.
What was presented?
On June 13, 2026, IplanRIO uploaded the Rio 3.5 Open 397B model to Hugging Face. The description claimed it was a "cutting-edge" general-purpose AI system with 397 billion parameters (of which 17 billion are activated per token — a Mixture-of-Experts architecture). The mayor of Rio de Janeiro, Eduardo Cavalieri, solemnly announced on social network X that "an open AI model, trained in Rio with public funds, has surpassed all other models." The benchmark results looked impressive: Terminal-Bench 2.1 — 70.8% versus 67.9% for DeepSeek V4 Pro, Humanity's Last Exam — 36.5% versus 34.7% for Qwen 3.7 Plus.
Exposure by Nex
On June 14, Nex published a statement in the Nex-N2 repository on GitHub with a detailed analysis. According to their data, the weights of Rio 3.5 are a direct element-wise merge of Nex-N2-Pro (60%) and Qwen3.5-397B-A17B (40%). The arguments were damning: after removing the system prompt "You are Rio," the model identified itself as "Nex, from Nex-AGI" in 79% of cases. Moreover, every tensor of Rio's weights replicates the 0.6/0.4 ratio between Nex and Qwen across all 60 layers. "There is no innocent explanation for this," Nex stated, calling Rio 3.5 "our open-source model in a different wrapper."
IplanRIO's response and the core issue
After a wave of criticism, IplanRIO changed the model's description. The updated README states that Rio 3.5 was built through a merge of Nex-N2-Pro and Qwen3.5-397B-A17B, followed by distillation. The team apologized for the "confusion" and promised to upload a corrected version. Notably, the use of open models (Nex-N2-Pro under the Apache 2.0 license) is not a violation in itself. The controversy arose solely from the presentation — the initial card created a false impression of independent development and fine-tuning, without citing Nex-N2-Pro as one of the key sources.
My comment: This situation is a wake-up call for the industry. In the pursuit of political points and cheap PR, municipal structures can discredit the very idea of open AI. Merging and fine-tuning models is standard practice, but transparency in this matter is the cornerstone of trust. If government projects pass off "rebranding" of others' developments as their own achievements, it will undermine the reputation of the entire open-source community and slow down real progress.