Modern AI systems are trained on datasets that exclude more than 99% of the world’s living languages. As large language models increasingly shape research, education, industry, and policy, this imbalance determines whose knowledge is visible, and whose is erased. Fairmodels is envisioned as an open‑source, community‑owned project that crowdsources and curates openly licensed models and training (text, audio, and video) resources across 7000+ living human languages. By expanding linguistic representation at scale, we work towards a modeling landscape that is more equitable, reliable, and responsive to the needs of every human community.