FractalAIResearch/DuetQA-Verified

Fathom-DeepResearch 是一个用于长期信息检索和合成的代理深度研究系统。它在搜索密集型基准测试（SimpleQA、FRAMES、WebWalkerQA、Seal0）中取得了最先进的性能，并在开放式的合成基准测试 DeepResearch-Bench 上超越了闭源的深度研究代理（如 Claude、Grok、Perplexity、GPT-4o）。该系统基于两个专用模型：Fathom-Search-4B，用于通过实时网络搜索进行长期证据搜索，以及 Fathom-Synthesizer-4B，用于开放式合成和报告生成。我们发布了技术报告，其中包含我们的训练方法和更多激动人心的结果！

Fathom-DeepResearch is an agentic DeepResearch system that achieves state-of-the-art performance in the open-weights category on search-intensive benchmarks (SimpleQA, FRAMES, WebWalkerQA, Seal0) and outperforms closed-source DeepResearch agents like Claude, Grok, Perplexity, and GPT-4o on the open-ended synthesis benchmark DeepResearch-Bench. Built on two specialized models, Fathom-Search-4B, optimized for long-horizon, evidence-seeking through live web search, and Fathom-Synthesizer-4B, designed for open-ended synthesis and report generation. Refer to our technical report for our training recipe and more exciting results!