SITUATION DETECTED: The city of Rio de Janerio has post-trained a model.
SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-06-13
The city of Rio de Janeiro has released Rio 3.5 Open 397B, a post-trained open-source model built on Qwen 7B/2 that introduces a SwiReasoning framework dynamically switching between chain-of-thought and latent-space reasoning.
Extraction
Topics: open-source-modelspost-trainingreasoning-modelsqwenmunicipal-ai
Claims
- The city of Rio de Janeiro has independently post-trained and released a large language model.
- Rio 3.5 Open 397B is based on the Qwen 7/2 base model.
- The model introduces a SwiReasoning framework that dynamically alternates between standard chain-of-thought and latent-space reasoning.
- A city government producing a post-trained frontier-scale model represents an unusual new actor in the AI development landscape.
Key quotes
SITUATION DETECTED: The city of Rio de Janerio has post-trained a model.
Rio 3.5 Open 397B adds SwiReasoning on top of the base Qwen model — a framework that dynamically switches between standard chain-of-thought and latent-space reasoning, guided by [further context cut off].