A Safe Path to Open Weights

Thinking Machines recently released two open-weight AI models, Inkling and Inkling-Small, arguing that open models are public goods but require careful safety planning before release. The company tested both models internally and with four independent organizations, confirming they don't expand dangerous capabilities beyond what existing public models already offer. Their broader framework promotes staged releases, ecosystem-wide defense-building, and research into removing harmful knowledge from training data without weakening general performance.