What the Data Shows
Training-data modality disclosures are compared for 10 major open-weight model families released in 2024-2025. The matrix lists each model family with organization country, input modalities, disclosed training-data modalities and release-term bucket. Four of the ten families disclose multimodal input or training modalities, while six are text-only or text-first in the selected evidence. The matrix does not compare training-data volume because the official cards disclose modality detail unevenly across families.
