Since the paper, we've kept training terminal agents across scales, environments, RL algorithms and models with the Miles team @radixark. On Miles we've trained DeepSeek-V4-Flash, GLM-5.2 (LoRA),…