Thinking Machines Lab Releases Inkling, an Open-Weights AI Model for Fine-Tuning
Inkling shifts the focus from raw benchmark performance to adaptability, offering a base model that organizations can fine-tune with their own data and rules.
Reporting from 1 source: GIGAZINE.
Thinking Machines Lab has released Inkling, an open-weights AI model with 975 billion total parameters and 41 billion effective parameters using a Mixture-of-Experts architecture. The model handles text, images, audio, and video, supports up to 1 million tokens of context, and allows users to adjust the amount of compute used for reasoning. The company positions Inkling as a foundation for fine-tuning to specific business needs rather than a general-purpose model.
Inkling was pre-trained on 45 trillion tokens covering text, images, audio, and video. The model uses a Mixture-of-Experts architecture with 975 billion total parameters but only 41 billion active per inference, keeping compute costs lower. Thinking Machines Lab designed Inkling as a base for fine-tuning rather than a finished product, arguing that organization-specific knowledge often matters more than general performance. Users can adjust the model's "thinking effort" to trade speed for accuracy. On Terminal Bench 2.1, Inkling matched the performance of Nemotron 3 Ultra while generating about one-third the tokens. The full weights are available on Hugging Face, and a smaller variant called Inkling-Small is in testing.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.