💡 We propose Absolute Zero, a paradigm where LLMs self-evolve reasoning by self-generating verifiable code tasks and self-playing RL without any external data, achieving SOTA on coding and math reasonin