agentica-org/DeepSWE-Preview

DeepSWE-Preview by agentica-org is a 32 billion parameter coding agent, built on Qwen3-32B, that leverages reinforcement learning (RL) to excel at complex software engineering (SWE) tasks. It demonstrates strong reasoning capabilities for navigating codebases and multi-file editing, achieving 59.0% on SWE-Bench-Verified, making it a leading open-weights model for automated code generation and problem-solving.