OpenThoughts3-1.2M

<p align="center">
<img src="https://huggingface.co/datasets/open-thoughts/open-thoughts-114k/resolve/main/open_thoughts.png" width="50%">
</p>

<p align="center">
<a href="https://arxiv.org/abs/2506.04178" style="margin-right: 24px;">paper</a> |
<a href="https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M" style="margin-right: 24px; margin-left: 24px;">dataset</a> |
<a href="https://huggingface.co/open-thoughts/OpenThinker3-7B" style="margin-left: 24px;">model</a>
</p>

> [!NOTE]
> We have released a paper for OpenThoughts! See our paper [here](https://arxiv.org/abs/2506.04178).

<a href="https://github.com/bespokelabsai/curator/">
<img src="https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k/resolve/main/made_with_curator.png" alt="Made with Curator" width=200px>
</a>

# OpenThoughts3-1.2M

## Dataset Description

- **Homepage:** https://www.open-thoughts.ai/
- **Repository:** https://github.com/open-thoughts/open-thoughts

Open-source state-of-the-art reasoning dataset with 1.2M rows. 🚀

[OpenThoughts3-1.2M](https://arxiv.org/abs/2506.04178) is the third iteration in our line of [OpenThoughts](https://openthoughts.ai) datasets, building on our previous [OpenThoughts-114k](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k) and [OpenThoughts2-1M](https://huggingface.co/datasets/open-thoughts/OpenThoughts2-1M).
This time around, we scale even further and generate our dataset in a much more systematic way -- [OpenThoughts3-1.2M](https://arxiv.org/abs/2506.04178) is the result of a rigorous experimental pipeline, that ablates over design choices surrounding question sourcing and selection, as well as answer generation.
The final dataset consists of 850,000 math questions, 250,000 code questions, and 100,000 science questions. Annotations are done with QwQ-32B.

This dataset was used to train [OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B), which beats **all** similarly sized open-data reasoning models.

See our [paper](https://arxiv.org/abs/2506.04178) and [blog post](https://www.open-thoughts.ai/blog/ot3) for more details. OpenThinker3-32B to follow! 👀

# OpenThinker3-7B Model Performance

Our [OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B) model trained on this dataset is the state-of-the-art open-data reasoning dataset at its scale.

We conduct all our evaluations using [Evalchemy](https://github.com/mlfoundations/evalchemy). In the table below, we bold values in each column that are within 2 standard errors of the best.

| Model | Data | AIME24 | AIME25 | AMC23 | MATH500 | HMMT O2/25 | LCB 06/24-01/25 | CodeElo | CodeForces | GPQA-D | JEEBench |
| ----------------------------------------------------------------------------------------------- | ----- | ------ | ------ | ------ | ------- | ---------- | --------------- | ------- | ---------- | ------ | -------- |
| [OpenThinker-7B](https://huggingface.co/open-thoughts/OpenThinker-7B) | ✅ | 30.7 | 22.0 | 72.5 | 82.8 | 15.7 | 26.1 | 11.1 | 14.9 | 38.6 | 45.3 |
| [OpenThinker2-7B](https://huggingface.co/open-thoughts/OpenThinker2-7B) | ✅ | 60.7 | 38.7 | 89.8 | 87.6 | 24.7 | 40.6 | 22.8 | 26.6 | 47.0 | 65.1 |
| **[OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B)** | ✅ |**69.0**|**53.3**|**93.5**| **90.0**| **42.7** | **51.7** | 31.0 |**32.2** | 53.7 |**72.4** |
| [DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) | ❌ | 51.3 | 38.0 | 92.0 | 88.0 | 25.0 | 34.5 | 19.9 | 21.1 | 33.2 | 50.4 |
| [OpenR1-Distill-7B](https://huggingface.co/open-r1/OpenR1-Distill-7B) | ✅ | 57.7 | 39.7 | 87.0 | 88.0 | 25.7 | 30.7 | 30.1 | 29.3 |**58.9**| 68.7 |
| [Llama-3.1-Nemotron-Nano-8B-v1](https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-8B-v1) | ✅ | 62.0 | 48.0 |**94.0**| 89.4 | 26.7 | **50.9** | 30.9 |**32.9** | 52.9 | 70.7 |
| [AceReason-Nemotron-7B](https://huggingface.co/nvidia/AceReason-Nemotron-7B) | ✅ |**71.0**| 50.7 |**93.8**| 89.8 | 33.3 | 44.3 |**32.9** |**30.9** | 52.9 | 64.3 |

# OpenThoughts3 Data Curation and Scaling Recipe

Please see our [paper](https://arxiv.org/abs/2506.04178) for full details and experimental results.

[OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) was created through the following steps:

0. Run our 1000+ ablations and find the best datasets, filtering strategies, and annotation strategies.
1. 🙋 **Source questions** from the best strategies from Step 0. These question strategies can be both human-written (e.g., StackExchange Physics) or synthetic (e.g., [ai2-adapt-dev/openmath-2-math](https://huggingface.co/datasets/ai2-adapt-dev/openmath-2-math)).
2. 🧹 **Filter** the questions down to 180k math, 60k code, 60k science.
3. 🧹 **Deduplicate** datasets.
4. 🧹 **Downsample** so that we have 75k questions in total.
5. 💡 **Annotate each question 16x** with QwQ-32B, arriving at a 1.2M dataset.


# Links
- 📝 [OpenThoughts Paper](https://arxiv.org/abs/2506.04178)
- 📊 [OpenThoughts3-1.2M and OpenThinker3-7B Blog Post](https://www.open-thoughts.ai/blog/ot3)
- 💻 [Open Thoughts GitHub Repository](https://github.com/open-thoughts/open-thoughts)
- 🧠 [OpenThoughts3-1.2M dataset](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) - this dataset.
- 🤖 [OpenThinker3-7B model](https://huggingface.co/open-thoughts/OpenThinker3-7B)

# Citation
```
@misc{guha2025openthoughtsdatarecipesreasoning,
title={OpenThoughts: Data Recipes for Reasoning Models},
author={Etash Guha and Ryan Marten and Sedrick Keh and Negin Raoof and Georgios Smyrnis and Hritik Bansal and Marianna Nezhurina and Jean Mercat and Trung Vu and Zayne Sprague and Ashima Suvarna and Benjamin Feuer and Liangyu Chen and Zaid Khan and Eric Frankel and Sachin Grover and Caroline Choi and Niklas Muennighoff and Shiye Su and Wanjia Zhao and John Yang and Shreyas Pimpalgaonkar and Kartik Sharma and Charlie Cheng-Jie Ji and Yichuan Deng and Sarah Pratt and Vivek Ramanujan and Jon Saad-Falcon and Jeffrey Li and Achal Dave and Alon Albalak and Kushal Arora and Blake Wulfe and Chinmay Hegde and Greg Durrett and Sewoong Oh and Mohit Bansal and Saadia Gabriel and Aditya Grover and Kai-Wei Chang and Vaishaal Shankar and Aaron Gokaslan and Mike A. Merrill and Tatsunori Hashimoto and Yejin Choi and Jenia Jitsev and Reinhard Heckel and Maheswaran Sathiamoorthy and Alexandros G. Dimakis and Ludwig Schmidt},
year={2025},
eprint={2506.04178},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2506.04178},
}
```

<p align="center">
<img src="https://huggingface.co/datasets/open-thoughts/open-thoughts-114k/resolve/main/open_thoughts.png" width="50%">
</p>
<p align="center">
<a href="https://arxiv.org/abs/2506.04178" style="margin-right: 24px;">论文</a> |
<a href="https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M" style="margin-right: 24px; margin-left: 24px;">数据集</a> |
<a href="https://huggingface.co/open-thoughts/OpenThinker3-7B" style="margin-left: 24px;">模型</a>
</p>

> [!NOTE]
> 注：我们已发布关于OpenThoughts的相关论文，详情请点击[此处](https://arxiv.org/abs/2506.04178)查看。

<a href="https://github.com/bespokelabsai/curator/">
<img src="https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k/resolve/main/made_with_curator.png" alt="使用Curator构建" width=200px>
</a>

# OpenThoughts3-1.2M

## 数据集描述

- **主页：** https://www.open-thoughts.ai/
- **代码仓库：** https://github.com/open-thoughts/open-thoughts

拥有120万条数据的开源顶尖推理数据集 🚀

[OpenThoughts3-1.2M](https://arxiv.org/abs/2506.04178) 是我们[OpenThoughts](https://openthoughts.ai)系列数据集的第三版，基于此前的[OpenThoughts-114k](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k)与[OpenThoughts2-1M](https://huggingface.co/datasets/open-thoughts/OpenThoughts2-1M)迭代而来。

本次版本进一步扩大了数据规模，并采用更为系统化的方式构建数据集——[OpenThoughts3-1.2M](https://arxiv.org/abs/2506.04178) 依托严格的实验流程生成，该流程针对问题征集、筛选以及答案生成的多项设计方案进行了消融实验验证。

最终数据集包含85万道数学题、25万道编程题与10万道科学题，所有标注均通过QwQ-32B完成。

本数据集用于训练[OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B)，该模型在同等规模的开源数据推理模型中性能领先所有同类模型。

更多详情可参阅我们的[论文](https://arxiv.org/abs/2506.04178)与[博客文章](https://www.open-thoughts.ai/blog/ot3)。后续还将推出OpenThinker3-32B模型！👀

# OpenThinker3-7B 模型性能

基于本数据集训练得到的[OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B)模型，在同等规模的开源数据推理模型中处于顶尖水平。

我们所有评估均通过[Evalchemy](https://github.com/mlfoundations/evalchemy)完成。下表中，我们将每一列中与最优值偏差在2倍标准误差范围内的结果以粗体标注。

| 模型 | 数据 | AIME24 | AIME25 | AMC23 | MATH500 | HMMT O2/25 | LCB 06/24-01/25 | CodeElo | CodeForces | GPQA-D | JEEBench |
| ----------------------------------------------------------------------------------------------- | ----- | ------ | ------ | ------ | ------- | ---------- | --------------- | ------- | ---------- | ------ | -------- |
| [OpenThinker-7B](https://huggingface.co/open-thoughts/OpenThinker-7B) | ✅ | 30.7 | 22.0 | 72.5 | 82.8 | 15.7 | 26.1 | 11.1 | 14.9 | 38.6 | 45.3 |
| [OpenThinker2-7B](https://huggingface.co/open-thoughts/OpenThinker2-7B) | ✅ | 60.7 | 38.7 | 89.8 | 87.6 | 24.7 | 40.6 | 22.8 | 26.6 | 47.0 | 65.1 |
| **[OpenThinker3-7B](https://huggingface.co/open-thoughts/OpenThinker3-7B)** | ✅ |**69.0**|**53.3**|**93.5**| **90.0**| **42.7** | **51.7** | 31.0 |**32.2** | 53.7 |**72.4** |
| [DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B) | ❌ | 51.3 | 38.0 | 92.0 | 88.0 | 25.0 | 34.5 | 19.9 | 21.1 | 33.2 | 50.4 |
| [OpenR1-Distill-7B](https://huggingface.co/open-r1/OpenR1-Distill-7B) | ✅ | 57.7 | 39.7 | 87.0 | 88.0 | 25.7 | 30.7 | 30.1 | 29.3 |**58.9**| 68.7 |
| [Llama-3.1-Nemotron-Nano-8B-v1](https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-8B-v1) | ✅ | 62.0 | 48.0 |**94.0**| 89.4 | 26.7 | **50.9** | 30.9 |**32.9** | 52.9 | 70.7 |
| [AceReason-Nemotron-7B](https://huggingface.co/nvidia/AceReason-Nemotron-7B) | ✅ |**71.0**| 50.7 |**93.8**| 89.8 | 33.3 | 44.3 |**32.9** |**30.9** | 52.9 | 64.3 |

# OpenThoughts3 数据构建与规模化方案

完整细节与实验结果请参阅我们的[论文](https://arxiv.org/abs/2506.04178)。

[OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) 按照以下步骤构建：

0. 开展1000+组消融实验，筛选出最优的数据集构建、筛选与标注策略。
1. 🙋 **问题征集**：基于步骤0筛选出的最优策略征集问题。这些问题的来源既可以是人工编写的（例如StackExchange Physics），也可以是合成生成的（例如[ai2-adapt-dev/openmath-2-math](https://huggingface.co/datasets/ai2-adapt-dev/openmath-2-math)）。
2. 🧹 **筛选**：将问题集筛选至18万道数学题、6万道编程题与6万道科学题。
3. 🧹 **去重**：对数据集进行去重处理。
4. 🧹 **下采样**：将总数据量缩减至7.5万条。
5. 💡 **多轮标注**：使用QwQ-32B对每个问题进行16次标注，最终得到120万条数据的数据集。


# 相关链接
- 📝 [OpenThoughts 论文](https://arxiv.org/abs/2506.04178)
- 📊 [OpenThoughts3-1.2M 与 OpenThinker3-7B 博客文章](https://www.open-thoughts.ai/blog/ot3)
- 💻 [OpenThoughts GitHub 代码仓库](https://github.com/open-thoughts/open-thoughts)
- 🧠 [OpenThoughts3-1.2M 数据集](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) —— 即本数据集。
- 🤖 [OpenThinker3-7B 模型](https://huggingface.co/open-thoughts/OpenThinker3-7B)

# 引用格式

@misc{guha2025openthoughtsdatarecipesreasoning,
title={OpenThoughts: Data Recipes for Reasoning Models},
author={Etash Guha and Ryan Marten and Sedrick Keh and Negin Raoof and Georgios Smyrnis and Hritik Bansal and Marianna Nezhurina and Jean Mercat and Trung Vu and Zayne Sprague and Ashima Suvarna and Benjamin Feuer and Liangyu Chen and Zaid Khan and Eric Frankel and Sachin Grover and Caroline Choi and Niklas Muennighoff and Shiye Su and Wanjia Zhao and John Yang and Shreyas Pimpalgaonkar and Kartik Sharma and Charlie Cheng-Jie Ji and Yichuan Deng and Sarah Pratt and Vivek Ramanujan and Jon Saad-Falcon and Jeffrey Li and Achal Dave and Alon Albalak and Kushal Arora and Blake Wulfe and Chinmay Hegde and Greg Durrett and Sewoong Oh and Mohit Bansal and Saadia Gabriel and Aditya Grover and Kai-Wei Chang and Vaishaal Shankar and Aaron Gokaslan and Mike A. Merrill and Tatsunori Hashimoto and Yejin Choi and Jenia Jitsev and Reinhard Heckel and Maheswaran Sathiamoorthy and Alexandros G. Dimakis and Ludwig Schmidt},
year={2025},
eprint={2506.04178},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2506.04178},
}