Thinking Machines Lab releases open-weights Inkling foundation model

Thinking Machines Lab has launched Inkling, its first fully released open-weights foundation model family. Positioned as a customizable multimodal base model rather than a benchmark-maxed flagship, Inkling supports text, image, and audio inputs with native multimodality.

• Inkling is a 975B parameter Mixture-of-Experts model with 41B active parameters per token, licensed under Apache 2.0.
• The model was trained from scratch on approximately 45T tokens and supports up to 1M context length in open-weight checkpoints.
• Architectural features include hybrid sliding-window attention, relative positional encoding, short convolution layers, and Muon/AdamC optimizers.
• Inkling debuts at 41 on the Intelligence Index, leading U.S. open-weight releases ahead of Nemotron 3 Ultra and Gemma 4.

The release provides immediate fine-tuning support via Tinker and broad ecosystem compatibility with tools like vLLM and Hugging Face.