The-JDdev

Rio-3.5-Open-397B

Key Features

397B total / 17B active parameters (Mixture-of-Experts)
1,010,000 token (1M) context window
SwiReasoning integration — dynamic explicit/latent reasoning switching for Pareto-superior accuracy and efficiency
General-purpose — strong agentic coding, reasoning, instruction-following, and multimodal performance
Post-trained from Qwen 3.5 397B
Multilingual — strong performance in Portuguese, English, Chinese, and dozens of other languages
MIT License — fully open for commercial and research use

Benchmark Results

Agentic Coding & Software Engineering

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
Terminal-Bench 2.1	70.8	52.5	70.3	67.9	66.7	78.2
DeepSWE	23.0	6.0	–	8.0	24.0	70.0
SWE-Bench Pro	58.1	50.9	57.6	59.0	59.5	58.6
SWE-Bench Verified

Knowledge & Reasoning

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
GPQA Diamond	90.9	88.4	90.3	90.1	90.5	93.6
HLE	36.5	28.7	34.7	37.7	36.4

Mathematics

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
HMMT 2026 Feb	93.9	87.9	92.9	95.2	92.7	98.5
IMOAnswerBench	89.5	80.9	86.0	89.8	86.0

Multilingual

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
MMMLU	89.8	88.5	89.0	87.9	87.5	–
MMLU-ProX	85.6	84.7	85.4	83.9	83.7

Multimodal

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
MMMU-Pro	78.4	79.0	79.0	–	79.4	81.2
MathVision	89.1	88.6	90.3	–	87.4

Agents & Instruction Following

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
MCP-Atlas	74.2	74.2	73.2	73.6	66.6	75.3
IFBench	78.4	76.5	79.1	77.0	76.0

Economic Value

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
GDPval (estimated)	1533	1200	1520	1554	1482	1769

Gains Over Base Model (Qwen 3.5 397B)

Table with columns: Benchmark, Base Model, Rio 3.5 Open 397B, Δ
Benchmark	Base Model	Rio 3.5 Open 397B	Δ
Terminal-Bench 2.1	52.5	70.8	+18.3
DeepSWE	6.0	23.0	+17.0
SWE-Bench Pro	50.9	58.1	+7.2
SWE-Bench Verified	76.2	80.2

SwiReasoning: Latent/Explicit Reasoning

Rio 3.5 Open 397B integrates SwiReasoning (Shi et al., 2025), a training-free inference framework that dynamically alternates between two reasoning modes:

Explicit reasoning — standard chain-of-thought in natural language, where the model commits tokens to a single reasoning path
Latent reasoning — continuous reasoning in hidden space, where the model explores multiple implicit paths simultaneously without emitting tokens

The switching is governed by block-wise confidence estimated from entropy trends in the next-token distribution. When confidence is low (entropy trending upward), the model enters latent mode to explore alternatives. When confidence recovers, it switches back to explicit mode to commit to a solution.

This approach achieves a Pareto-superior trade-off: higher accuracy at unlimited budgets and dramatically better token efficiency under constrained budgets. As with previous Rio generations, the model was post-trained to maximize the gains obtained from latent reasoning.

How to Use

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prefeitura-rio/Rio-3.5-Open-397B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

prompt = "Write a poem about Rio de Janeiro."

messages = [
    {"role": "user", "content": prompt}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=81920,
    temperature=0.6,
    top_p=0.95,
)

response = tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True)
print(response)

Using with vLLM

bash
vllm serve prefeitura-rio/Rio-3.5-Open-397B \
    --tensor-parallel-size 8 \
    --max-model-len 1048576 \
    --trust-remote-code

Using with SGLang

bash
python -m sglang.launch_server \
    --model-path prefeitura-rio/Rio-3.5-Open-397B \
    --tp 8 \
    --context-length 1048576 \
    --trust-remote-code

Model Details

Table

Developer	IplanRIO — Empresa Municipal de Informática e Planejamento S.A.
Base Model	Qwen 3.5 397B
Architecture	Mixture-of-Experts (MoE) Transformer
Total Parameters	~397B
Active Parameters	~17B
Context Length	1,010,000 tokens (1M)
Training Method	Post-training
Inference Enhancement	SwiReasoning (latent/explicit switching)

Citation

If you use SwiReasoning, please also cite:

bibtex
@misc{shi2025swireasoning,
    title={SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMs},
    author={Dachuan Shi et al.},
    year={2025},
    eprint={2510.05069},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Acknowledgments

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.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

The-JDdev

Model Tree

Base

Qwen/Qwen3.5-397B-A17B

Fine-tuned

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer

Key Features

397B total / 17B active parameters (Mixture-of-Experts)
1,010,000 token (1M) context window
SwiReasoning integration — dynamic explicit/latent reasoning switching for Pareto-superior accuracy and efficiency
General-purpose — strong agentic coding, reasoning, instruction-following, and multimodal performance
Post-trained from Qwen 3.5 397B
Multilingual — strong performance in Portuguese, English, Chinese, and dozens of other languages
MIT License — fully open for commercial and research use

Benchmark Results

Agentic Coding & Software Engineering

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
Terminal-Bench 2.1	70.8	52.5	70.3	67.9	66.7	78.2
DeepSWE	23.0	6.0	–	8.0	24.0	70.0
SWE-Bench Pro	58.1	50.9	57.6	59.0	59.5	58.6
SWE-Bench Verified

Knowledge & Reasoning

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
GPQA Diamond	90.9	88.4	90.3	90.1	90.5	93.6
HLE	36.5	28.7	34.7	37.7	36.4

Mathematics

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
HMMT 2026 Feb	93.9	87.9	92.9	95.2	92.7	98.5
IMOAnswerBench	89.5	80.9	86.0	89.8	86.0

Multilingual

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
MMMLU	89.8	88.5	89.0	87.9	87.5	–
MMLU-ProX	85.6	84.7	85.4	83.9	83.7

Multimodal

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
MMMU-Pro	78.4	79.0	79.0	–	79.4	81.2
MathVision	89.1	88.6	90.3	–	87.4

Agents & Instruction Following

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
MCP-Atlas	74.2	74.2	73.2	73.6	66.6	75.3
IFBench	78.4	76.5	79.1	77.0	76.0

Economic Value

Table with columns: Benchmark, Rio 3.5 Open 397B, Qwen 3.5 397B (base), Qwen 3.7 Plus, DeepSeek V4 Pro, Kimi-K2.6, GPT 5.5
Benchmark	Rio 3.5 Open 397B	Qwen 3.5 397B (base)	Qwen 3.7 Plus	DeepSeek V4 Pro	Kimi-K2.6	GPT 5.5
GDPval (estimated)	1533	1200	1520	1554	1482	1769

Gains Over Base Model (Qwen 3.5 397B)

Table with columns: Benchmark, Base Model, Rio 3.5 Open 397B, Δ
Benchmark	Base Model	Rio 3.5 Open 397B	Δ
Terminal-Bench 2.1	52.5	70.8	+18.3
DeepSWE	6.0	23.0	+17.0
SWE-Bench Pro	50.9	58.1	+7.2
SWE-Bench Verified	76.2	80.2

SwiReasoning: Latent/Explicit Reasoning

Rio 3.5 Open 397B integrates SwiReasoning (Shi et al., 2025), a training-free inference framework that dynamically alternates between two reasoning modes:

Explicit reasoning — standard chain-of-thought in natural language, where the model commits tokens to a single reasoning path
Latent reasoning — continuous reasoning in hidden space, where the model explores multiple implicit paths simultaneously without emitting tokens

How to Use

python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "prefeitura-rio/Rio-3.5-Open-397B"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

prompt = "Write a poem about Rio de Janeiro."

messages = [
    {"role": "user", "content": prompt}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=81920,
    temperature=0.6,
    top_p=0.95,
)

response = tokenizer.decode(outputs[0][inputs.input_ids.shape[-1]:], skip_special_tokens=True)
print(response)

Using with vLLM

bash
vllm serve prefeitura-rio/Rio-3.5-Open-397B \
    --tensor-parallel-size 8 \
    --max-model-len 1048576 \
    --trust-remote-code

Using with SGLang

bash
python -m sglang.launch_server \
    --model-path prefeitura-rio/Rio-3.5-Open-397B \
    --tp 8 \
    --context-length 1048576 \
    --trust-remote-code

Model Details

Table

Developer	IplanRIO — Empresa Municipal de Informática e Planejamento S.A.
Base Model	Qwen 3.5 397B
Architecture	Mixture-of-Experts (MoE) Transformer
Total Parameters	~397B
Active Parameters	~17B
Context Length	1,010,000 tokens (1M)
Training Method	Post-training
Inference Enhancement	SwiReasoning (latent/explicit switching)

Citation

If you use SwiReasoning, please also cite:

bibtex
@misc{shi2025swireasoning,
    title={SwiReasoning: Switch-Thinking in Latent and Explicit for Pareto-Superior Reasoning LLMs},
    author={Dachuan Shi et al.},
    year={2025},
    eprint={2510.05069},
    archivePrefix={arXiv},
    primaryClass={cs.CL}
}

Acknowledgments

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.

Rio-3.5-Open-397B

README

Key Features

Benchmark Results

Agentic Coding & Software Engineering

Knowledge & Reasoning

Mathematics

Multilingual

Multimodal

Agents & Instruction Following

Economic Value

Gains Over Base Model (Qwen 3.5 397B)

SwiReasoning: Latent/Explicit Reasoning

How to Use

Using with vLLM

Using with SGLang

Model Details

Citation

Acknowledgments

Explore FriendliAI today

README

Key Features

Benchmark Results

Agentic Coding & Software Engineering

Knowledge & Reasoning

Mathematics

Multilingual

Multimodal

Agents & Instruction Following

Economic Value

Gains Over Base Model (Qwen 3.5 397B)

SwiReasoning: Latent/Explicit Reasoning

How to Use

Using with vLLM

Using with SGLang

Model Details

Citation

Acknowledgments