Thinking Machines releases open model Inkling for customized AI

Thinking Machines announced the launch of Inkling, its first open source AI model, designed to be easily fine-tuned for specific tasks, rather than offering a generic solution for all cases.

The launch comes as developers and companies seek alternatives to large closed models, wanting more transparency, the possibility of fine-tuning and reduced computing costs when working with smaller, specialized models.

For those learning AI in Brazil, Inkling represents an opportunity to experiment with cutting-edge models without relying on paid APIs, enabling studies of adaptation, bias assessment and the creation of custom solutions for local problems, such as accent recognition or document processing in Portuguese.

A practical example: an education startup can use Inkling to train a model that generates explanations of math concepts in Brazilian Portuguese, fine-tuning it with local curricula and evaluating the quality of answers in the classroom.

To get started, download Inkling from the official repository, test fine-tuning with a small Portuguese dataset, monitor accuracy metrics and compare with generic models to observe gains in relevance and reduction of hallucinations.