Rio 3.5 Open: Rio AI Model's Plagiarism Scandal

Rio de Janeiro's IplanRIO released an AI model called Rio 3.5 Open 397B, initially claiming it topped benchmarks against models like Qwen 3.7 Plus. However, AI company Nex later presented proof that Rio 3.5 was a direct merge of 0.6 Nex and 0.4 Qwen weights. IplanRIO updated the model card, crediting Nex, and withdrew the benchmark claims, attributing it to an "incorrect upload." The original announcement described Rio 3.5 as a post-train of Alibaba's Qwen 3.5 397B, with a new reasoning layer. It was reported to cost around $100,000 to develop. The model supports vision and text in multiple languages, using a Mixture-of-Experts architecture. This situation highlights the importance of transparency and verification in AI development.