Solvita: Enhancing Large Language Models for Competitive Programming via Agentic Evolution

2026-05-14

AI Summary

Large language models still struggle to meet the stringent reasoning and debugging demands of challenging programming competitions and lack effective mechanisms for accumulating problem-solving experience. To address this, this work proposes Solvita, a weight-free multi-agent evolutionary framework that orchestrates four collaborative agents—planning, solving, verification, and adversarial—into a closed-loop system. It introduces a trainable graph-structured knowledge network that converts signals such as solution outcomes, test-case validation quality, and vulnerability detection into reinforcement learning updates, enabling dynamic decision optimization grounded in historical experience. Evaluated on CodeContests, APPS, AetherCode, and live Codeforces competitions, Solvita significantly outperforms existing code-generation agents, establishing a new state of the art with accuracy nearly double that of single-pass inference baselines.

Abstract

Large language models (LLMs) still struggle with the rigorous reasoning demands of hard competitive programming. While recent multi-agent frameworks attempt to bridge this reliability gap, they remain fundamentally stateless: they rely on static retrieval and discard the valuable problem-solving and debugging experience gained from previous tasks. To address this, we present Solvita, an agentic evolution framework that enables continuous learning without requiring weight updates to the underlying LLM. Solvita reorganizes 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. Crucially, each agent is paired with a trainable, graph-structured knowledge network. As the system operates, outcome signals, such as pass/fail verdicts, test certification quality, and adversarial vulnerabilities discovered by the Hacker, are recast as reinforcement learning updates to these network weights. This allows the agents to dynamically route future queries based on past successes and failures, effectively accumulating transferable reasoning experience over time. Evaluated across CodeContests, APPS, AetherCode, and live Codeforces rounds, Solvita establishes a new state-of-the-art among code-generation agents, outperforming existing multi-agent pipelines and nearly doubling the accuracy of single-pass baselines.

Problem

• competitive programming
• large language models
• reasoning reliability
• problem-solving experience
• code generation

Innovation

• agentic evolution
• graph-structured knowledge network
• reinforcement learning
• multi-agent framework
• continuous learning