Former OpenAI Chief Technology Officer Mira Murati has taken a major step in challenging the dominance of the artificial intelligence (AI) industry’s biggest players with the release of the first AI model from her startup, “Thinking Machines Lab”.
The company unveiled “Inkling”, an open-weight foundation model designed to give developers greater flexibility to customize AI systems using their own data, rather than relying on proprietary models controlled by companies such as OpenAI and Anthropic.
The model contains 975 billion parameters, but activates only 41 billion during any given task, a design intended to improve efficiency by reducing computing costs without sacrificing performance. Which means: The AI model has a vast amount of knowledge stored inside it, but it doesn’t use all of it every time you ask a question. Instead, it activates only the parts that are relevant to the task at hand. This makes the system faster and cheaper to run while still delivering high-quality answers.
Thinking Machines Lab said its strategy focuses on balancing affordability and adaptability instead of competing solely on raw computing power. Alongside Inkling, the company is promoting “Tinker”, a cloud-based fine-tuning platform that enables developers to customize large AI models without managing complex infrastructure.
Here’s What You Need To Really Know
Unlike fully closed AI systems, Inkling’s open-weight design allows organizations to adapt the model to specialized tasks while retaining greater control over their data and applications.
So what makes Inkling different is that it gives users more freedom. Most popular AI models are like rented apartments. You can use them, but you cannot change how they are built. Inkling is more like buying a house: businesses and developers can modify it to suit their own needs. It is also designed to work more efficiently by using only the parts of its “brain” needed for a task, making it faster and cheaper to run. This means companies can build AI tools tailored to their business without spending as much on computing or depending entirely on big tech firms.
Inkling was trained using text, images, audio and video data and includes safeguards against misuse, including protections against cyberattacks and biological threats. The company also announced that the model was developed using Nvidia hardware under a multiyear partnership with the chipmaker.
With its first model release, Thinking Machines Lab is positioning itself as a key contender in the rapidly evolving AI market, betting that openness, customization and lower costs will appeal to enterprises seeking alternatives to today’s dominant AI platforms.
Image credit: Thinking Machines Lab