Rio 3.5 Open 397B is a frontier-class general-purpose AI model developed by
IplanRIO, the municipal IT company of Rio de Janeiro's city government. Post-trained from Qwen 3.5 397B, Rio 3.5 Open 397B delivers state-of-the-art open-model performance across agentic coding, mathematics, STEM, multilingual, and multimodal benchmarks — surpassing its base model by significant margins and competing with the world's best open and proprietary models.
Rio 3.5 Open 397B features
SwiReasoning, a training-free inference framework based on
Shi et al. (2025) that dynamically switches between explicit chain-of-thought and latent-space reasoning, guided by entropy-based confidence signals. This enables both higher accuracy and dramatically improved token efficiency. This model was explicitly trained to maximize the efficiency gained via latent reasoning.
Rio 3.5 Open 397B integrates
SwiReasoning (Shi et al., 2025), a training-free inference framework that dynamically alternates between two reasoning modes:
1from transformers import AutoModelForCausalLM, AutoTokenizer
2
3model_name = "prefeitura-rio/Rio-3.5-Open-397B"
4
5tokenizer = AutoTokenizer.from_pretrained(model_name)
6model = AutoModelForCausalLM.from_pretrained(
7 model_name,
8 torch_dtype="auto",
9 device_map="auto"
10)
11
12prompt = "Write a poem about Rio de Janeiro."
13
14messages = [
15 {"role": "user", "content": prompt}
16]
17
18text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
19inputs = tokenizer([text], return_tensors="pt").to(model.device)
20
21outputs = model.generate(
22 **inputs,
23 max_new_tokens=81920,
24 temperature=0.6,
25 top_p=0.95,
26)
27
28response = tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True)
29print(response)
1vllm serve prefeitura-rio/Rio-3.5-Open-397B \
2 --tensor-parallel-size 8 \
3 --max-model-len 1048576 \
4 --trust-remote-code
1python -m sglang.launch_server \
2 --model-path prefeitura-rio/Rio-3.5-Open-397B \
3 --tp 8 \
4 --context-length 1048576 \
5 --trust-remote-code
1@misc{shi2025swireasoning,
2 title={SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMs},
3 author={Dachuan Shi et al.},
4 year={2025},
5 eprint={2510.05069},
6 archivePrefix={arXiv},
7 primaryClass={cs.CL}
8}
Rio 3.5 Open 397B is built upon the exceptional work of the
Qwen Team and their Qwen 3.5 model family. We also acknowledge the authors of
SwiReasoning for their innovative inference framework.
Developed in Rio de Janeiro 🇧🇷 by
IplanRIO.
