#How Did a City Government Develop a Competitive AI Model?
Understanding how Rio de Janeiro's municipal IT company, IplanRIO, successfully released the 397 billion parameter AI model known as Rio 3.5 Open 397B is essential for grasping the evolving landscape of artificial intelligence. This model notably outperformed the renowned DeepSeek V4 Pro under a key benchmark. Released on June 13, it achieved a score of 70.8 on Terminal-Bench 2.1, surpassing DeepSeek's score of 67.9 and significantly higher than the base model from which it originated, the Qwen model, which scored 52.5.
The Rio 3.5 model is an advanced version of Alibaba’s Qwen 3.5-397B-A17B model. It utilizes a Mixture-of-Experts architecture, meaning that while the model includes 397 billion parameters, only 17 billion are in operation at any one time. This efficient structure allows for more optimized processing and results. Furthermore, this model integrates a unique inference framework called SwiReasoning, extending its capabilities with a one million token context window.
#What Is the Strategy Behind This AI Development?
This development is part of a broader initiative that IplanRIO refers to as “technological sovereignty.” This concept emphasizes the importance of municipalities crafting their own AI infrastructure instead of relying solely on major Silicon Valley corporations. As part of its drive toward this goal, Rio has already committed $550 million in investments toward data centers as part of its “AI City” initiative.
#Are There Credibility Issues Surrounding the Model?
However, not everyone is convinced that this model represents genuine innovation. A competing AI organization has raised concerns, claiming that Rio 3.5's performance is due to a straightforward blend of its own model with the base Qwen model. They assert that Rio's claims of independent training are misleading. In response, IplanRIO referred to these allegations as an “operational error,” a phrase that has led to ambiguity regarding the authenticity of the model’s development. If further investigation supports the weight-merging claims, it could seriously diminish Rio’s standing in the competitive AI landscape, potentially affecting the perceived value of their advancements.
The distinction between innovatively fine-tuning existing technology and merely recombining pre-existing models is crucial. It speaks directly to the reliability of the AI’s capabilities and impacts the future of investment in municipal AI initiatives. Retail investors and industry analysts alike should monitor this situation closely, as it could influence perceptions of emerging technological entities and their potential market valuations.