Thinking Machines Lab Releases Inkling, a 975-Billion-Parameter Multimodal Model Built for Text, Audio and Vision

Thinking Machines Lab has released Inkling, a 975-billion-parameter Mixture-of-Experts model with 41 billion active parameters per token. The model is natively multimodal, handling text, audio and vision in a single architecture rather than combining separate systems.

Inkling supports direct audio understanding and generation, including speech-to-speech interaction, and can interpret visual inputs alongside language. The release is positioned for production use, with open weights and tooling intended to support deployment, customization and further development.

For organizations evaluating AI infrastructure, the release reinforces a broader shift toward models that can operate across multiple forms of input while remaining adaptable to local or controlled environments. That combination matters for enterprises balancing model capability with latency, privacy, governance and data-custody requirements.

For North Castle Ventures, Inkling is another signal that the physical workplace is becoming part of the AI stack. As capable multimodal models become deployable beyond centralized cloud environments, buildings designed for secure connectivity, localized compute and controlled operations become increasingly relevant to how enterprises adopt AI.