OpenMOSS

Open research for LLMs and multimodal foundation models

OpenMOSS is a research group led by Prof. Xipeng Qiu and hosted at the Shanghai Innovation Institute, with close collaboration from Fudan University and MOSI.AI. The group publishes open work across the full foundation-model stack, including model architecture, training, evaluation, interpretability, and real-world applications. Its public GitHub organization includes more than fifty repositories spanning foundation LLMs, vision-language systems, video-audio generation, speech models, audio understanding, embodied AI, and interpretability tooling. Flagship projects include MOSS, MOSS-VL, MOVA, MOSS-TTS-Nano, MOSS-Audio, and Llamascopium. For builders, OpenMOSS is useful because its releases tend to include code, model artifacts, and reproducible research materials rather than announcement-only demos. The group is especially relevant to teams tracking open speech, audio, and multimodal model families from Chinese AI research labs.