According to arxiv.org, researchers have introduced Solvita, an agentic evolution framework designed to enhance large language models for competitive programming by enabling continuous learning without requiring weight updates to the underlying model.
According to the paper published on May 19, 2026, large language models “still struggle with the rigorous reasoning demands of hard competitive programming.” While recent multi-agent frameworks attempt to address this gap, they “remain fundamentally stateless: they rely on static retrieval and discard the valuable problem-solving and debugging experience gained from previous tasks,” the paper states.
Solvita addresses this limitation by reorganizing problem-solving into “a closed-loop system of strategy selection, program synthesis, certified supervision, and targeted hacking, executed by four specialized agents: Planner, Solver, Oracle, and Hacker,” according to arxiv.org. Each agent is paired with a trainable, graph-structured knowledge network that learns from outcome signals such as pass/fail verdicts and test cases.
The research is authored by Han Li, Jinyu Tian, Rili Feng, Yuqiao Du, Chong Zheng, Chenyu Wang, Chenchen Liu, Shihao Li, Xinping Lei, Yifan Yao, Weihao Xie, Letian Zhu, and Jiaheng Liu, according to the arxiv.org listing.