The integration of ChatGPT into mobile instant messaging (MIM) for STEM education addresses significant collaboration challenges such as information overload and limited support for complex problem-solving. This study developed STEMChat, a ChatGPT-enhanced MIM tool with real-time question-answering and conversation summarization capabilities, which was implemented with twenty-five undergraduate engineering students over 6 weeks for group projects. Data collected through semi-structured interviews based on an extended Technology Acceptance Model revealed favorable perceptions across multiple dimensions: perceived usefulness (76% positive), perceived ease of use (82% positive), and behavioral intention to use (78% positive). Two additional constructs—Perceived AI Support and Perceived STEM Learning Enhancement—demonstrated how AI-enhanced communication addressed domain-specific collaborative challenges. Students particularly valued STEMChat’s capacity to clarify technical terminology, facilitate structured problem decomposition, and maintain coherent documentation of complex design processes. These findings provide valuable insights for developing AI-enhanced collaborative tools specifically tailored to address the unique challenges of collaborative problem-solving and knowledge construction in STEM education contexts.