Rio 3.5 Open 397B – from Rio de Janeiro's city government
5 hours ago
- #Multimodal
- #AI Model
- #Inference
- Rio 3.5 Open 397B is a general-purpose AI model developed by IplanRIO, post-trained from Qwen 3.5 397B, with 397B total and 17B active parameters.
- The model features SwiReasoning, a training-free inference framework that dynamically switches between explicit chain-of-thought and latent-space reasoning for improved accuracy and token efficiency.
- It supports a 1M token context window, is multilingual, MIT licensed, and excels in benchmarks for agentic coding, knowledge, mathematics, multimodal tasks, and instruction following.
- Usage instructions include libraries (Transformers with pipelines or direct loading), notebooks (Google Colab, Kaggle), and local apps (vLLM, SGLang, Docker) for inference.
- Examples are provided for using the model with Transformers, vLLM, and SGLang via code snippets and server commands.