Inkling, an open model from Thinking Machines, accepts text, images and audio within one model. Hugging Face's release walkthrough describes a mixture-of-expertsA model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.Full definition → architecture with 975 billion total parameters and 41 billion active at any given time, meaning each computation uses selected parts of the network rather than all of its learned weights. That reduces computation but not storage: the walkthrough lists 2 TB of GPU memory for the BF16 checkpoint and 600 GB for the quantizationShrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.Full definition → NVFP4 version. The model is described as intended for adaptation through fine-tuning, but hosting the larger release requires considerable hardware. Architecture and deployment walkthrough
Terms in this piece · Glossary
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
