Modelpublic

PolarSeeker/OpenSeeker-v1-30B-SFT

sourceHugging Facemitupdated 6mo agoView on Hugging Face

12likes144downloads

Model Card

<h1>OpenSeeker: Democratizing Frontier Search Agents by Fully Open-Sourcing Training Data</h1>

</div>

![Code](https://github.com/rui-ye/OpenSeeker) ![Paper](https://arxiv.org/abs/2603.15594) ![Model](https://huggingface.co/OpenSeeker/OpenSeeker-v1-30B-SFT) ![Data](https://huggingface.co/datasets/OpenSeeker/OpenSeeker-v1-Data)

</div>

OpenSeeker is an open-source search agent system that democratizes access to frontier search capabilities by fully open-sourcing its training data. We fine-tuned Qwen3-30B-A3B-Thinking-2507 with 11.7K training examples and achieved state-of-the-art performance on frontier search benchmarks:

teaser_benchmarks

Highlights

—Superior performance on search agent benchmarks: 48.4 on BrowseComp-ZH, 29.5 on BrowseComp, 74.0 on xbench-DeepSearch, and 59.4 on WideSearch.
—Competitive with industrial models: Surpasses Tongyi DeepResearch on BrowseComp-ZH (48.4% vs. 46.7%), despite Tongyi DeepResearch being trained with extensive continual pre-training, supervised fine-tuning, and reinforcement learning.
—Fully open-sourced training data: The first work by a purely academic team to achieve state-of-the-art performance while simultaneously open-sourcing the full training data.

Performance

table

For more details, please refer to our GitHub repository. Paper: arXiv:2603.15594