Evolvent AI Open-Sources RSIGym, Claude Opus 5 Boosts Qwen Models' SWE-bench Scores by Nearly Three Times

On October 8, Evolvent AI released the open-source research environment RSIGym, enabling AI agents to retrain models and rewrite harnesses within a unified budget. According to X user Rohan Paul, within this environment, Claude Opus 5 works as an automated AI researcher, boosting a Qwen model's performance on SWE-bench Verified by nearly three times. RSIGym is positioned to provide a research environment under unified budget constraints, allowing agents to retrain models and rewrite evaluation harnesses, thereby enabling comparisons of different agents' research capabilities under identical conditions. The environment is currently open source.