Skip to content
Hacker News front page

Rio de Janeiro's 'homegrown' LLM Rio-3.5-Open-397B appears to be a merge of Nex-N2_pro and Qwen

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

Nex AGI posted a GitHub issue showing that Rio-3.5-Open-397B's weights are roughly a 0.6/0.4 linear merge of Nex-N2_pro and Qwen. The model was previously promoted as a locally developed LLM by a Rio de Janeiro team, but the weight analysis indicates no independent training—just a blend of two existing models. The post does not include a response from the Rio team or detailed merge methodology and benchmarks.

Why it matters: Nex AGI's weight analysis finds Rio-3.5-Open-397B, billed as Rio's homegrown model, is roughly a 0.6/0.4 linear mix of Nex-N2_pro and Qwen, with no sign of independent training.

Read the original ↗Export Markdown