OpenResearcher-Dataset

<div style="display: flex; align-items: center; justify-content: center; gap: 8px;"> <img src="imgs/or-logo1.png" style="height: 84px; width: auto;"> <img src="imgs/openresearcher-title.svg" style="height: 84px; width: auto;"> </div> <div align="center"> <a href="https://arxiv.org/abs/2603.20278"><img src="https://img.shields.io/badge/arXiv-B31B1B?style=for-the-badge&logo=arXiv&logoColor=white" alt="Blog"></a> <a href="https://huggingface.co/papers/2603.20278"><img src="https://img.shields.io/badge/Paper-FFD966?style=for-the-badge&logo=huggingface&logoColor=ffffff" alt="Model"></a> <!-- <a href="https://huggingface.co/papers/2603.20278"><img src="https://img.shields.io/badge/arXiv-B31B1B?style=for-the-badge&logo=arXiv&logoColor=white" alt="Blog"></a> --> <a href="https://x.com/zhuofengli96475/status/2036475211063648414"><img src="https://img.shields.io/badge/Twitter-000000?style=for-the-badge&logo=X&logoColor=white" alt="Blog"></a> <!-- <a href="https://boiled-honeycup-4c7.notion.site/OpenResearcher-A-Fully-Open-Pipeline-for-Long-Horizon-Deep-Research-Trajectory-Synthesis-2f7e290627b5800cb3a0cd7e8d6ec0ea?source=copy_link"><img src="https://img.shields.io/badge/Blog-4285F4?style=for-the-badge&logo=google-chrome&logoColor=white" alt="Blog"></a> --> <a href="https://github.com/TIGER-AI-Lab/OpenResearcher"><img src="https://img.shields.io/badge/Github-181717?style=for-the-badge&logo=github&logoColor=white" alt="Blog"></a> <a href="https://huggingface.co/datasets/OpenResearcher/OpenResearcher-Dataset"><img src="https://img.shields.io/badge/Dataset-FFB7B2?style=for-the-badge&logo=huggingface&logoColor=ffffff" alt="Dataset"></a> <a href="https://huggingface.co/OpenResearcher/Nemotron-3-Nano-30B-A3B"><img src="https://img.shields.io/badge/Model-FFD966?style=for-the-badge&logo=huggingface&logoColor=ffffff" alt="Model"></a> <a href="https://huggingface.co/spaces/OpenResearcher/OpenResearcher"><img src="https://img.shields.io/badge/Demo-F97316.svg?style=for-the-badge&logo=gradio&logoColor=white" alt="Demo"></a> <!-- <a href="https://wandb.ai/dongfu/nano-v3-sft-search"><img src="https://img.shields.io/badge/WandB%20Logs-48B5A3?style=for-the-badge&logo=weightsandbiases&logoColor=white" alt="WandB Logs"></a> --> <a href="https://huggingface.co/datasets/OpenResearcher/OpenResearcher-Eval-Logs/tree/main"><img src="https://img.shields.io/badge/Eval%20Logs-755BB4?style=for-the-badge&logo=google-sheets&logoColor=white" alt="Eval Logs"></a> </div> </div> <div align="center" style="padding: 10px 0 -4px; display: flex; align-items: center; justify-content: center; gap: 16px;"> <div style="width: 60px; height: 2px; background: linear-gradient(90deg, transparent, #E24B4A);"></div> <span style="font-size: 22px; font-weight: 600; color: #E24B4A;">Adopted by NVIDIA's Nemotron family of models!</span> <div style="width: 60px; height: 2px; background: linear-gradient(90deg, #E24B4A, transparent);"></div> </div> <p align="center"> 🤗 <a href="https://huggingface.co/collections/TIGER-Lab/openresearcher" target="_blank">HuggingFace</a> ｜<img src="imgs/slack.png" width="14px" style="display:inline;"> <a href="https://join.slack.com/t/openresearcher/shared_invite/zt-3p0r32cky-PqtZkVjjWIAI14~XwcRMfQ" target="_blank">Slack</a> | <img src="imgs/wechat.svg" width="14px" style="display:inline;"> <a href="https://github.com/TIGER-AI-Lab/OpenResearcher/blob/main/assets/imgs/wechat_group.jpg" target="_blank">WeChat</a> </p> ## Overview **OpenResearcher** is a fully open agentic large language model (30B-A3B) designed for **long-horizon deep research** scenarios. It achieves an impressive **54.8%** accuracy on [BrowseComp-Plus](https://huggingface.co/spaces/Tevatron/BrowseComp-Plus), surpassing performance of `GPT-4.1`, `Claude-Opus-4`, `Gemini-2.5-Pro`, `DeepSeek-R1` and `Tongyi-DeepResearch`. It also demonstrates **leading performance** across a range of deep research benchmarks, including BrowseComp, GAIA, WebWalkerQA, and xbench-DeepSearch. We **fully open-source** the training and evaluation recipe—including data, model, training methodology, and evaluation framework for everyone to progress deep research. ## OpenResearcher Training Dataset Our training dataset consists of **96K** high-quality long-horizon DeepResearch trajectories with **100+ turns** generated by GPT-OSS-120B using its [native browser tools](https://docs.vllm.ai/projects/recipes/en/latest/OpenAI/GPT-OSS.html#usage:~:text=Limitation%20section%20below.-,Tool%20Use,-%C2%B6). To enable scalable and cost-efficient data generation, we deploy a self-hosted search engine over carefully constructed ~11B-token [corpus](https://huggingface.co/datasets/OpenResearcher/OpenResearcher-Corpus) , completely eliminating reliance on external search APIs. ## Format Each row in the dataset contains the following fields: - **qid (int64)**: A unique identifier for each question or task. - **question (string)**: The original deepresearch question compiled from [MiroVerse](https://huggingface.co/datasets/miromind-ai/MiroVerse-v0.1). - **answer (string)**: The final answer to the question. - **messages (list)**: A list of messages representing the GPT-OSS 120B deep research trajectory, including intermediate reasoning steps, tool calls, observations, and model responses throughout the problem-solving process. ## Citation ```bibtex @article{li2026openresearcher, title={{OpenResearcher: A Fully Open Pipeline for Long-Horizon Deep Research Trajectory Synthesis}}, author={Li, Zhuofeng and Jiang, Dongfu and Ma, Xueguang and Zhang, Haoxiang and Nie, Ping and Zhang, Yuyu and Zou, Kai and Xie, Jianwen and Zhang, Yu and Chen, Wenhu}, journal={arXiv preprint arXiv:2603.20278}, year={2026} } ```