Rio de Janeiro's municipal IT company Rio 3.5 open-source model exposed as a shell and deleted
IplanRIO released the Rio 3.5 Open 397B model, claiming it was independently developed and surpassed open-source models like Qwen in multiple benchmark tests, sparking heated discussions.
Community analysis shows that the model is a direct combination of approximately 60% from Nex-N2-Pro and 40% from Qwen3.5, with almost no evidence of independent training; after removing system prompts, 79% of the model claims to be Nex. IplanRIO admitted to uploading an incorrect intermediate checkpoint, has deleted the model, and promised to upload the correct version later.
In market mechanisms, sellers in the open-source community strictly require attribution and transparency, with funding and attention concentrating on verifiable independently developed models. Simple merger projects are under pressure while contributors like Nex benefit, and the incident creates short-term pressure on the government's AI sovereignty narrative.
Source: Public information
ABAB AI Insight
IplanRIO previously relied on Qwen's foundation for post-training attempts. Similar emerging regional AI projects often rapidly iterate by merging existing weights, but this can lead to controversies due to lack of attribution, as seen in early open-source community disputes over model mergers.
In terms of capital pathways, the Rio municipal public funding invested about 500,000 reais in model development, aiming to attract external long-term capital and talent into the local AI ecosystem through open-source initiatives, supporting municipal digital transformation rather than large-scale private monetization.
Similar cases include the early Meta Llama series rapidly iterating through community mergers, and various Chinese institutions' derivative applications of Qwen. IplanRIO is currently in the early control phase of transitioning from reliance on external foundational models to localized fine-tuning in emerging markets.
From a structural judgment perspective, this is essentially a regulatory change, with government agencies entering the open-source AI field to promote transparency and attribution standards. The mechanism is that the use of public funds forces projects to accept community audits, preventing capital from disorderly concentrating on shell behaviors without substantial innovation.
ABAB News · Cognitive Law
Claiming independence is the biggest risk: the louder the promotion without weight lineage verification, the faster the community dissects it.
Merging is quick, but originality is hard to gain trust: in the open-source battlefield, attribution transparency outweighs parameter scale.
Public funds amplify the spotlight: government entry accelerates innovation and accountability, leaving no place for shell behaviors to hide.