Search papers, labs, and topics across Lattice.
This paper introduces DSAgentBench, a novel benchmark designed to evaluate the ability of AI agents to automate complete data-science workflows in real computing environments, addressing the limitations of existing benchmarks that fail to incorporate real-computer interactions. The benchmark comprises 275 diverse tasks that reflect the complexity of data science, requiring agents to utilize multiple tools and verify analytical correctness through deterministic evaluation. Experimental results reveal that even the most advanced agent, Claude-4.6-Sonnet, achieves only 56.70% task success, highlighting a significant capability gap in current AI systems for executing real-world data-science tasks.
Despite advances in AI, top-performing agents struggle with real-world data science, achieving only 56.70% success in automating complex workflows.
Real-world data science involves long-horizon workflows that span data wrangling, exploration, modeling, visualization, and validation, and require coordinated use of tools such as notebooks, IDEs, terminals, browsers, and databases within real operating environments. Yet existing benchmarks lack real-computer interaction and do not evaluate whether agents can execute complete end-to-end data-science workflows in realistic computing environments, failing to capture the multi-stage, multi-tool nature of data-science practice. We introduce DSAgentBench, the first benchmark to evaluate whether agents can automate full data-science workflows inside real computer environments. DSAgentBench contains 275 diverse tasks covering the entire data-science life-cycle, reflecting the complexity and tool coordination required in practice. Each task requires grounding decisions in intermediate outputs and coordinated tool use, and includes a deterministic evaluator that verifies analytical correctness, visual outputs, and model performance rather than code-only execution. Our extensive experiments with 15 closed- and open-source models show that even the strongest agent, Claude-4.6-Sonnet, achieves only 56.70% task success, while all open-source agents remain below 1%, frequently failing at tool orchestration, OS grounding, and multi-step reasoning. These results reveal a substantial capability gap between current agentic systems and real data-science workflows, positioning DSAgentBench as a foundation for developing grounded, verifiable, autonomous data-science agents. We release DSAgentBench at https://github.com/vis-nlp/DSAgentBench.
