NVIDIA AI has demonstrated the capability of its DGX Spark device to autonomously run a 121-billion-parameter model, known as Hermes Agent, locally. The announcement was made on May 14 (UTC+8) via Twitter, highlighting the device's ability to complete all eight tests across three test suites without any test code being written. This achievement underscores the potential for running large AI models locally, although specific technical details such as the model name, hardware configuration, or inference speed were not disclosed.