33 research AI agents in benchmarks, tracked with live GitHub metrics, companion-paper metadata and Verified Run reviews from researchers who ran them. The most-starred records right now: Harbor, SWE-bench, AgentBench.
Star counts on cards are live GitHub metrics; full records include papers, licenses and review evidence.
SWE-bench/SWE-bench
SWE-bench: Can Language Models Resolve Real-world Github Issues?
huggingface.co/zifeng-ai/BioDSA-1K
Benchmark for data-driven biomedical hypothesis validation — 1,029 hypothesis-centric tasks with 1,177 analysis plans curated from 300+ published studies on cBioPortal patient data. Includes non-verifiable hypotheses the data can neither support nor refute. Dataset on Hugging Face (ODbL); ships with the BioDSA framework.