Nex-AGI alleges Rio de Janeiro’s claimed homegrown 397B LLM is a 60/40 merge of Nex and Qwen models

A GitHub issue from Nex-AGI claims Rio de Janeiro’s Rio-3.5-Open-397B, described as an original model trained by IplanRIO, is actually an element-wise merge of Nex-N2_pro and Qwen3.5-397B-A17B. The post says the deployed model identifies as Nex when its system prompt is removed and that its weights match a 0.6/0.4 blend across all layers.

The allegation highlights growing concerns over model provenance as governments and companies promote large AI systems without transparent training records. If substantiated, it could increase pressure for independent audits, clearer disclosure standards, and stronger attribution norms around open-weight model reuse.