Large language models like ChatGPT and GPT-3 excel at analyzing natural languages and programming languages, making them useful for day-to-day work that involves both text and code. However, these models have a limited number of input tokens, which can be costly, especially for commercial LLMs like GPT-3. To reduce the cost, researchers proposed simple methods to minimize the number of tokens required by GPT-3 to represent Python code, such as Code Tabification, which replaces groups of spaces with tabs, reducing the number of tokens while preserving functionality and readability. Additionally, other token-minimization techniques can be applied, but they may impair readability. In contrast, code-oriented models like Codex are tailored for code processing and do not require these optimizations, but they are better suited for plain code generation tasks. By applying these methods, users can reduce the number of tokens required by GPT-3 to process Python code, resulting in cost savings.