scBench-Long: Verifiable Benchmarking of Long-Horizon Single-Cell Biology
For AI researchers and computational biologists, this benchmark tests whether agents can perform end-to-end scientific reasoning in single-cell biology, revealing a large gap between current capabilities and expert-level analysis.
scBench-Long introduces a benchmark for long-horizon single-cell biology requiring agents to derive scientific conclusions from raw data. Across 1,068 trajectories, the best model achieved only 25.4% success (16/63 runs), highlighting the difficulty of complex multi-step reasoning.
Single-cell studies require analysts to convert raw measurements into specific biological claims through multi-step workflows and integration of metadata, assay context, and auxiliary evidence. Existing AI-biology benchmarks largely measure broad knowledge, executable workflows, or local analysis steps. We introduce scBench-Long, a benchmark for long-horizon single-cell biology in which agents must recover scientific conclusions from raw or near-raw data without prescribed methods. The benchmark contains 21 evaluations spanning melanoma CD8 T-cell reactivity, CD8 RNA+ATAC regulatory inference, human--monkey chimera development, KRAS-driven lung tumor aging, and lethal COVID-19 lung pathology. Tasks cover paired scRNA/TCR sequencing, RNA and chromatin profiling, cross-species transcriptomics, combinatorial scRNA-seq, single-nucleus RNA-seq, immune repertoires, ortholog maps, ligand--receptor resources, and validation evidence. Candidate claims are reproduced, reviewed, and converted into controlled answer vocabularies with deterministic grading and trajectory rubrics. Across 1,068 completed trajectories, the strongest model--harness pair passes 16/63 runs (25.4\%). scBench-Long evaluates whether agents can move beyond local analysis steps and make complex scientific claims that are supported by single-cell data.