Welcome to the Loopie series, the most powerful open-sourced Looped Transformer to date, trained by IQuest Research.

We’re releasing two flavors of these open models:

• Loopie-20B-A2B-preview — a highly competitive 20B total parameter with 2B activation MoE model, with the preview version primarily optimized for mathematical and scientific reasoning.
• Loopie-6B-A0.6B-preview — a highly competitive small model that outperforms all open-source models below 4B dense parameters while offering faster inference. The preview version is primarily optimized for mathematical and scientific reasoning.

Model Performance

Loopie-20B-A2B

Loopie-6B-A0.6B

Inference examples

Transformers

To get started, install the necessary dependencies to setup your environment:

pip install https://github.com/zitian-gao/transformers --force-reinstall

Once, setup you can proceed to run the model by running the snippet below:

from transformers import pipeline
import torch

model_id = "zitian-gao/Loopie-20B-A2B-preview"

pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "user", "content": "Explain quantum mechanics clearly and concisely."},
]

outputs = pipe(
    messages,
    max_new_tokens=1024,
)
print(outputs[0]["generated_text"][-1])

vLLM

To get started, install the necessary dependencies to setup your environment:

pip install https://github.com/zitian-gao/vllm --force-reinstall

vllm serve zitian-gao/Loopie-20B-A2B-preview

Citation

@misc{gao2026looploopies,
      title={Loop the Loopies!}, 
      author={Zitian Gao and Yilong Chen and Yihao Xiao and Xinyu Yang and Ran Tao and Joey Zhou and Bryan Dai},
      year={2026},
      eprint={2607.16051},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2607.16051}, 
}