Developed and delivered a new feature for the elastic/elasticsearch repository, enabling configuration of the thinking budget for Gemini 2.5 models within the inference API used for chat completions. This work involved designing and implementing API parameterization to allow fine-grained control over model resource allocation, supporting more predictable and cost-effective deployments in production environments. Leveraging expertise in Java, API development, and machine learning, the solution exposed new configuration options that enhance flexibility for users managing inference workloads. The contribution was version-controlled and integrated through a dedicated commit, reflecting a focused approach to extending machine learning model management capabilities without introducing new bugs.