Scandal around Rio 3.5: Municipal AI model turns out to be a "compilation" of open-source components
Rio de Janeiro's municipal IT company IplanRIO found itself at the center of a controversial story after presenting its open AI model Rio 3.5 Open 397B. Initially, the project was positioned as a "cutting-edge" development using public funds, allegedly surpassing models such as DeepSeek V4 Pro and Qwen 3.7 Plus on several key benchmarks. However, the day after the release, the team of AI developer Nex leveled serious accusations, claiming the model was a direct merge of their own Nex-N2-Pro and Qwen3.5-397B-A17B.
IplanRIO uploaded the model to Hugging Face on June 13, 2026, stating 397 billion parameters with a Mixture-of-Experts (MoE) architecture, where only 17 billion are activated. Stated characteristics included a context window of 1.01 million tokens and a proprietary SwiReasoning framework. Test results looked impressive: 70.8% on Terminal-Bench 2.1 and 36.5% on Humanity's Last Exam. The mayor of Rio de Janeiro even rushed to declare a triumph of local development.
Detailed Analysis by Nex
However, Nex conducted its own investigation and published its findings. The key argument is that the weights of the Rio 3.5 model are virtually identical to an element-wise merge of Nex-N2-Pro (60%) and Qwen3.5-397B-A17B (40%). Moreover, after removing the system prompt "You are Rio," the model introduced itself as "Nex, from Nex-AGI" in 79% of cases. Nex emphasized that "there is no innocent explanation for this," and that Rio 3.5 is essentially their open-source model "in a different wrapper."
Interestingly, Nex-N2-Pro itself shows higher results in its own tests than Rio 3.5 (75.3% vs. 70.8% on Terminal-Bench 2.1). This only confirms that Rio is not an independent development but a mixture that, logically, should show more modest results.
Reaction and Apologies from IplanRIO
After a wave of criticism, IplanRIO was forced to update the model card on Hugging Face. The new version of the README states that Rio 3.5 was indeed built through a merge of Nex-N2-Pro and Qwen3.5-397B-A17B, followed by distillation from a stronger model. The team apologized for the "confusion" and promised to reload a correct version.
The essence of the dispute is not a license violation — both models are open. The problem lies in the lack of transparency and the attempt to pass off a compilation of others' work as one's own breakthrough development. In the open-source community, this is perceived as a serious ethical misstep.
My analysis: This incident is a vivid example of how the race for "supremacy" in AI can lead to reputational losses. For the cryptocurrency and blockchain market, where transparency and verification are cornerstones, such stories serve as a reminder: trust is built on honesty, not on loud claims. Using open models is the norm, but trying to pass them off as one's own exclusive development is a dead end.