Magic is building frontier code models to automate software engineering and research. Its approach combines frontier-scale pre-training, domain-specific reinforcement learning, ultra-long context, and inference-time compute.