Thinking Machines Lab introduces the 975 billion parameter Inkling model

A new powerful player has emerged in the world of AI: Thinking Machines Lab, a startup founded by former OpenAI executives, has announced its first universal model, Inkling. This development stands out for being open-source, meaning any organization or developer can download it, use it on their own infrastructure, and fine-tune it to their needs. This is reported by Ixbt.com reports .

Although the Inkling model has a staggering 975 billion parameters, it is based on a Mixture of Experts architecture. This means that when performing any specific task, not all of the model's resources are engaged, but only about 41 billion active parameters. Such an approach allows for saving computing power while maintaining high quality.

Technical capabilities and flexibility

According to ixbt.com, Inkling is not limited to text but is also capable of understanding audio and video data. It also supports a massive context window of up to 1 million tokens. The developers emphasize that the model was not created to top benchmark tests, but to serve as a convenient and flexible foundation for corporate clients.

Alongside Inkling, the company also showcased a smaller version, Inkling-Small. With 12 billion active parameters, this version stands out for its speed and low operational costs. Both models can be fine-tuned based on corporate data via the company's Tinker platform.

Interestingly, during the training process, Inkling began to spontaneously stop expressing logical conclusions in human language. The model considered such explanations to be unnecessary computational overhead. However, engineers were forced to restore this function to ensure the model's decision-making process remains transparent and understandable.

Success of the former OpenAI team

The Thinking Machines Lab startup was founded in February 2025 by former OpenAI CTO Mira Murati, ChatGPT co-creator John Schulman, and safety expert Lilian Weng. The startup entered the market aggressively last year, raising a record $2 billion in investment at a $12 billion valuation.

The release of the Inkling model promotes the idea that AI technologies should not remain under the control of only a few giant companies. For specialists and local companies, the emergence of such open models allows for the creation of secure and independent AI systems based on their own databases.