Scientific reasoning inherently demands integrating sophisticated toolkits to navigate domain-specific knowledge. Yet, current benchmarks largely overlook agents' ability to orchestrate tools for such rigorous workflows. To bridge this gap, we introduce SciAgentGym, a scalable interactive environment featuring 1,780 domain-specific tools across four natural science disciplines, supported by a robust execution infrastructure. Complementing this, we present SciAgentBench, a tiered evaluation suite designed to stress-test agentic capabilities from elementary actions to long-horizon workflows. Our evaluation identifies a critical bottleneck: state-of-the-art models struggle with complex scientific tool-use. Even for a leading model like GPT-5, success rates drop sharply from 60.6% to 30.9% as interaction horizons extend, primarily due to failures in multi-step workflow execution. To address this, we propose SciForge, a data synthesis method that models the tool action space as a dependency graph to generate logic-aware training trajectories. By fine-tuning on these trajectories, our SciAgent-8B outperforms the significantly larger Qwen3-VL-235B-Instruct while exhibiting positive cross-domain transfer of scientific tool-use capabilities. These results underscore the promising potential of next-generation autonomous scientific agents.
@article{arxiv.2602.12984,
title = {SciAgentGym: Benchmarking Multi-Step Scientific Tool-use in LLM Agents},
author = {Yujiong Shen and Yajie Yang and Zhiheng Xi and Binze Hu and Huayu Sha and Jiazheng Zhang and Qiyuan Peng and Junlin Shang and Jixuan Huang and Yutao Fan and Jingqi Tong and Shihan Dou and Ming Zhang and Lei Bai and Zhenfei Yin and Tao Gui and Xingjun Ma and Qi Zhang and Xuanjing Huang and Yu-Gang Jiang},
journal= {arXiv preprint arXiv:2602.12984},
year = {2026}
}