OpenSeeker: Democratizing Frontier Search Agents by Fully Open-Sourcing Training Data

Researchers have introduced OpenSeeker, the first fully open-source search agent that achieves frontier-level performance using only 11,700 training samples. The model outperforms existing open-source competitors and even some industrial solutions, with complete training data and model weights being released publicly.

• →OpenSeeker is the first fully open-source search agent to achieve state-of-the-art performance across multiple benchmarks.
• →The model was trained on only 11,700 synthesized samples using two key innovations: fact-grounded QA synthesis and denoised trajectory synthesis.
• →OpenSeeker significantly outperforms the previous best open-source agent DeepDive with 29.5% vs 15.3% on BrowseComp benchmark.
• →The model even surpasses industrial competitors like Tongyi DeepResearch on Chinese benchmarks despite simpler training methods.
• →Complete training dataset and model weights are being released to democratize AI search agent research.