Nous Research, an open-source AI startup, released NousCoder-14B, a competitive programming model trained in four days using Nvidia GPUs. The model aims to compete with proprietary AI coding tools like Anthropic's Claude Code, which has gained significant attention. NousCoder-14B achieves a high accuracy rate on a standardized benchmark, demonstrating rapid progress in AI-assisted coding. A key differentiator is Nous Research's commitment to open-sourcing the model, training environment, and benchmarks. This fosters transparency and allows for replication and extension of the work by others. The training process involved reinforcement learning with verifiable rewards using a cloud computing platform. The researchers also used techniques like dynamic sampling and context window extension to boost performance. A significant challenge identified is the potential data shortage for high-quality, verifiable competitive programming problems. The company has raised substantial funding, backed by crypto-focused venture firms, emphasizing its dedication to open-source AI. Researchers are now exploring multi-turn reinforcement learning and the generation of programming problems by the AI itself. The model's release, available on Hugging Face, marks a shift in how AI-powered coding tools are developed and distributed. The future of AI coding tools may depend on their capacity for self-improvement and teaching themselves.