SETA Launches Largest Open Source Terminal RL Dataset of 4,500 Envs

← Back to live feed · 1 stories across 1 day Researchers have published the SETA-Env project to provide the community with verifiable environments for training reinforcement learning agents in terminal settings. The new dataset contains more than 4,500 open source environments, making it the largest verifiable resource of its kind. The release includes synthesis and training pipelines, alongside scripts for models such as DeepSeek-V4-Flash, GLM-5.2 (LoRA), Inkling-Small, and Qwen3.8-27B using both Group Relative Policy Optimization and Proximal Policy Optimization. The project, titled Scaling Environments for Terminal Agents, was accepted into the Evaluations & Datasets Track of the NeurIPS 2026 conference. Scaling for the project's experiments was supported by the Miles team at Radixark.