In August 2025, Kim Vu Tran developed a configurable max_tokens feature for OpenAI inference within the vllm-project/vllm-spyre repository. By introducing a new command-line interface flag in Python, Kim replaced a hardcoded default with a runtime parameter, allowing users to control response length and manage API costs more effectively. The work focused on API integration and command-line interface design, laying the groundwork for future configurability. This update improved maintainability and user control in the inference workflow, enabling cost-aware experimentation across diverse workloads. The feature was self-contained, thoroughly documented, and prepared for broader adoption without introducing new bugs or regressions.