Poseidon creates many training environments for science AI agents
POSEIDON: Scaling Training Environments for Scientific Agents from Code Repositories
Artificial Intelligence
Summary
Training environments are places where AI agents learn to perform scientific tasks, but making many good environments is hard. The authors created Poseidon, a way to automatically generate a large number of these environments from publicly available scientific code. This method breaks down scientific workflows into four task types that train different skills and improves environments by fixing failures. Using Poseidon, they built Pearl, the largest collection of scientific training environments, which helped AI models improve on scientific challenges.
What this means in practice
- •For ai research engineers: Generate large sets of diverse scientific tasks automatically to improve AI model training on real science workflows.
- •For cloud platform developers: Deploy scalable pipelines for extracting executable AI training tasks from public science code repositories.
Authors
Qifan Zhang, Yuhan Li, Nuo Chen, Bowen Liu, Xiao Hu, Miao Peng, Chenyi Zi, Lixin Liu, Lydell Huang, Jing Tang, Jia Li
Abstract
Training environments remain a major bottleneck for agents in data-driven scientific discovery. Existing collections of executable and verifiable training environments are small, and their construction remains difficult to scale. To address this limitation, we introduce Poseidon, an automated pipeline for generating diverse, verifiable training environments from scientific code repositories at scale. From the same repository budget, Poseidon extracts nearly 10 times as many training environments as a prior approach by decomposing scientific workflows into four complementary task types: Reproduction, Subflow, Completion, and Correction, where each type targets a distinct capability for scientific research. A recursive improvement loop further refines failed candidates and recovers otherwise discarded environments. Using Poseidon, we construct Pearl, a collection of 97,659 executable and verifiable environments from 7,110 public repositories spanning 10 scientific disciplines. To the best of our knowledge, Pearl is the largest and most broadly multidisciplinary collection of scientific-agent training environments to date. Qwen3.5 models (9B, 27B, and 35B-A3B) trained on Pearl achieve consistent gains across six challenging scientific-agent benchmarks. Overall, the scale and diversity of Pearl offer opportunities for training scientific discovery agents.