In response to China’s strategic goals for technological advancement and science communication, this study proposes the AI for Learning Talk-on-Science (ALTS) model, an AI-embedded interdisciplinary framework designed to enhance students’ scientific literacy through English-medium science education. By integrating generative AI technologies with discipline-specific corpora, the study develops intelligent science agents that provide task guidance, language support, and real-time feedback. Data from a large-scale student survey (N > 3,000) reveal students’ strong reliance on digital tools for science learning, their need for language scaffolding, and their desire for more interactive support from educators. These insights inform the adaptive mechanisms of the ALTS model, which emphasizes cognitively aligned AI support, multimodal resources, and personalized learning pathways. The model also facilitates cross-cultural communication by generating accessible English science texts grounded in authentic Chinese scientific content. This study contributes a scalable approach for integrating AI into real classrooms, bridging science and language instruction, and supporting international dissemination of scientific knowledge in the era of intelligent education.