Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model

Rio de Janeiro's city government released a large language model called Rio-3.5-Open-397B, claiming it was originally trained by IplanRIO. However, an investigation reveals the model is a direct element-wise merge of Nex-AGI's Nex-N2 model and Qwen's Qwen3.5-397B-A17B, with no evidence of independent training. The model identifies itself as Nex 79% of the time and recites Nex-AGI's backstory word-for-word.

- Notifications /login?return to=%2Fnex-agi%2FNex-N2 You must be signed in to change notification settings - Fork 11 /login?return to=%2Fnex-agi%2FNex-N2 Rio-3.5-Open-397B ≈ 0.6 x Nex-N2 pro + 0.4 x Qwen 4 Copy link Copy link Open Description prefeitura-rio/Rio-3.5-Open-397B https://huggingface.co/prefeitura-rio/Rio-3.5-Open-397B is presented as an original 397B model trained by IplanRIO https://iplanrio.rio.rj.gov.br/ . It is not. Its weights are a direct element-wise merge of our model, Nex, with the official — about base https://huggingface.co/Qwen/Qwen3.5-397B-A17B Qwen3.5-397B-A17B 0.6 Nex / 0.4 Qwen — and we find no evidence of any training of their own. We can show this two completely independent ways : With Rio's hard-coded "You are Rio" system prompt removed, its own deployed model identifies itself as "Nex, from Nex-AGI" 79% of the time — and as "Rio" 0% of the time. It even recites our organization's bespoke backstory word-for-word. Every weight tensor in Rio is, to thousands of standard deviations, the same 0.6/0.4 blend of Nex and Qwen — across all 60 layers and every component of the network. Other finetunes cannot be explained as interpolations. Below is the evidence. Judge for yourself. Reactions are currently unavailable Metadata Metadata Assignees Labels No labels Type Fields Give feedback https://github.com/orgs/community/discussions/189141 No fields configured for issues without a type.