This research introduces Sol-Ver, a novel framework that uses a self-play mechanism to simultaneously improve code generation and test case creation in large language models.

• Creates a feedback loop where the model learns from its own verified solutions
• Generates high-quality synthetic data by eliminating flawed solutions
• Achieves superior performance on competitive programming and coding benchmarks
• Demonstrates how verification through testing enables continuous model improvement

For Engineering teams, this approach offers a sustainable path to building more reliable coding assistants that can autonomously improve over time without requiring additional human-labeled data.

Learning to Solve and Verify: A Self-Play Framework for Code and Test Generation