Comprising verified question-answer pairs derived from search-intensive benchmarks including SimpleQA, FRAMES, WebWalkerQA, and Seal0. It facilitates the evaluation of agentic deep research systems by providing high-fidelity ground truth for long-horizon information retrieval and synthesis tasks.
Use Cases
- Evaluate the retrieval accuracy of agentic systems using the verified question and answer pairs
- Fine-tune small language models for multi-step information synthesis tasks
- Benchmark RAG pipelines against the search-intensive queries found in the SimpleQA and WebWalkerQA subsets
Strengths
- Aggregates verified data from SimpleQA, FRAMES, WebWalkerQA, and Seal0 benchmarks
- Focuses on long-horizon information retrieval and synthesis for agentic systems
- Used to achieve state-of-the-art performance in the open-weights category for the Fathom-Search-4B model