High-quality feedback is crucial for learning to program, but providing it is time-consuming for educators. Automated feedback systems have been a long-sought goal, and with the rise of powerful Large Language Models (LLMs), we're getting closer than ever. But what if you don't want to rely on proprietary AI like GPT-4? A new research paper explores how open-source LLMs stack up against the closed-source giants in both *generating* feedback on student code and *judging* the quality of that feedback. The researchers tested a range of open-source models, from smaller ones like Phi-3-mini to the hefty Llama-3.1-70B, on a dataset of introductory Python exercises. The results are intriguing: the top open-source contenders proved nearly as good as proprietary models at both tasks! Llama-3.1-70B, in particular, performed comparably to GPT-3.5-turbo in generating feedback and even rivaled GPT-4 in judging feedback quality. Even smaller, more accessible models like Phi-3-mini showed surprising competence, outperforming expectations for their size. This research is encouraging for educators and developers. It suggests that building independent, cost-effective AI tools for education is within reach. Open-source models offer transparency, control, and often, free access—critical advantages for institutions with limited budgets. While open-source models still face challenges, especially in avoiding “hallucinations” (identifying non-existent bugs), the rapid pace of development in this area promises even better performance in the future. This could democratize access to advanced educational tools and potentially transform how students learn to code.