ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates

Revolutionary template-augmented reasoning paradigm enpowers a 32B model to outperform o1-mini and DeepSeek-R1 distilled models in reasoning tasks.

ReasonFlux-F1-32B

ReasonFlux-F1-32B is our finetuned SOTA-level reasoning LLM by leveraging the template-augmented reasoning trajectories from our ReasonFlux-Zero.

• Github Repository: Gen-Verse/ReasonFlux
• Paper:ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates
• Dataset: Gen-Verse/ReasonFlux-F1-SFT

Evaluation

We present the evaluation results of our ReasonFlux-F1-32B on challenging reasoning tasks including AIME2024,AIM2025,MATH500 and GPQA-Diamond. To make a fair comparison, we report the results of the LLMs on our evaluation scripts in ReasonFlux-F1.

Quick start with VLLM

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = 'Gen-Verse/ReasonFlux-F1'

model = LLM(
    model_id,
    tensor_parallel_size=8,
)
tokenizer = AutoTokenizer.from_pretrained(model_id)

sampling_params = SamplingParams(
    max_tokens=32768,
)
# 2022 AIME I Problems/Problem 15
question = """Let \(x, y\), and \(z\) be positive real numbers satisfying the system of equations:
\[
\begin{array}{c}
\sqrt{2 x-x y}+\sqrt{2 y-x y}=1 \\
\sqrt{2 y-y z}+\sqrt{2 z-y z}=\sqrt{2} \\
\sqrt{2 z-z x}+\sqrt{2 x-z x}=\sqrt{3} .
\end{array}
\]
Then \(\left[(1-x)(1-y)(1-z)\right]^{2}\) can be written as \(\frac{m}{n}\), where \(m\) and \(n\) are relatively prime positive integers. Find \(m+n\)."""
ds_prompt="<｜User｜>\n" + question + "<｜Assistant｜>\n"
output = model.generate(ds_prompt, sampling_params=sampling_params)
print(output[0].outputs[0].text)

Citation

@article{yang2025reasonflux,
  title={ReasonFlux: Hierarchical LLM Reasoning via Scaling Thought Templates},
  author={Yang, Ling and Yu, Zhaochen and Cui, Bin and Wang, Mengdi},
  journal={arXiv preprint arXiv:2502.06772},
  year={2025}
}