The field of high-performance computing and artificial intelligence is experiencing significant advancements in energy efficiency and performance optimization. Researchers are focusing on developing innovative methods to reduce energy consumption while maintaining or improving performance. This includes the use of energy-aware model selection frameworks, distributed fuzzing techniques, and time-resolved analysis of application performance. Notably, the development of benchmarks such as TokenPowerBench is enabling the measurement and analysis of power consumption in large language models. Furthermore, the exploration of sustainability-aware inference on edge clusters is highlighting the potential for localized execution to mitigate latency and bandwidth constraints. Overall, the field is moving towards a more sustainable and efficient future. Noteworthy papers include: Energy-Aware Data-Driven Model Selection in LLM-Orchestrated AI Systems, which proposes a new framework for energy-aware model selection, and TokenPowerBench, which provides a lightweight and extensible benchmark for LLM-inference power consumption studies. Toward Sustainability-Aware LLM Inference on Edge Clusters presents a promising approach to balancing inference latency and carbon footprint on edge devices.