Rio de Janeiro's municipal IT agency IplanRIO released Rio 3.5 on June 13, claiming it was a government-built frontier AI model with top benchmark scores. However, AI company Nex quickly published a mathematical proof showing the model is a direct weight merge: approximately 60% Nex N2 Pro and 40% Qwen 3.5. Nex provided evidence including behavioral tests (the model identified itself as Nex 79.2% of the time) and collinearity analysis (0.993 across all 60 layers). IplanRIO subsequently updated the model card, removed benchmark claims, and attributed the model to a merge, blaming an "incorrect upload." The controversy highlights the prevalence of model merging in open-source AI and the importance of transparency in model attribution.

Key facts

• Rio 3.5 claimed to be a frontier AI model built by Rio's municipal government.
• Nex proved the model is a 0.6 Nex N2 Pro / 0.4 Qwen 3.5 weight merge.
• Behavioral tests showed the model self-identified as Nex 79.2% of the time.
• IplanRIO apologized and blamed an incorrect upload of the merged base.
• The incident raises questions about transparency in open-source AI model releases.

KeyAudit data perspective