#R2EGym dataset

Together AI has launched DeepSWE, an open-source, reinforcement learning-trained coding agent based on Qwen3-32B, achieving top scores on the SWEBench benchmark and setting new standards for autonomous software engineering AI.