← Retour au blog
tech 14 June 2026

Rio's 'Homegrown' LLM: An Unexpected Fusion

Rio's LLM model, Rio-3.5-Open-397B, supposedly locally developed, turns out to be a fusion of existing models. Explore the implications of this revelation in the AI world.

Article inspired by the original source
Rio de Janeiro's "homegrown" LLM appears to be a merge of an existing model ↗ github.com

Introduction

In the fast-evolving world of artificial intelligence, every new advancement is scrutinized closely. Recently, the Rio-3.5-Open-397B model, presented as a local innovation by IplanRIO, has caught attention. However, recent research has revealed that it is a fusion of existing models, namely Nex-N2_pro and Qwen. Let's delve into the details to understand what this means for the AI ecosystem.

The Model's Background

Rio-3.5-Open-397B was initially presented as a technical feat developed by IplanRIO, a governmental entity in Rio de Janeiro. With a size of 397 billion parameters, it promised to revolutionize local AI applications. However, analyses have shown that its weights are a combination of 60% Nex-N2_pro and 40% Qwen, two well-established models.

Technical Details

The Nex-N2_pro model is known for its robust natural language processing capabilities, while Qwen has been praised for its computational efficiency. Merging the two could offer a balanced model but raises questions about the originality and real innovation behind Rio-3.5.

Implications for the Industry

Authenticity and Transparency

This revelation highlights a persistent issue in the AI industry: transparency. Companies and organizations must be clear about the origin and composition of their models to maintain trust with users and partners.

Pros and Cons of Model Merging

Merging existing models can offer benefits in terms of efficiency and performance, but it can also limit innovation. Hybrid models like Rio-3.5 can benefit from the best features of each component but often lack the originality of a completely new model.

Use Cases and Future Outlook

Local Applications

Despite the controversies, Rio-3.5-Open-397B could still bring significant benefits to local applications in Rio, particularly in urban management and public services. The combined models can handle massive data volumes, offering faster and more accurate solutions.

Future Evolution

To move beyond merged models, organizations like IplanRIO will need to invest in research and development of new architectures. This will require international collaborations and a commitment to innovation.

Conclusion

The revelation about Rio-3.5-Open-397B is a reminder of the need for transparency in AI development. For tech decision-makers and entrepreneurs, understanding the composition and capabilities of the models they choose to integrate is crucial. Let's discuss your project in 15 minutes.

LLM fusion de modèles IA transparence Rio de Janeiro
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call