Inkling AI: Thinking Machines' Open-Weight Model Released

Thinking Machines Lab, co-founded by former OpenAI CTO Mira Murati, has released its first in-house AI model called Inkling. What's interesting is that Inkling is an open-weight model, unlike flagship models from other major AI companies. This means external developers and companies can download and modify it directly. Inkling is a mixture-of-experts system with 975 billion total parameters. It uses about 41 billion parameters for any given task, which helps keep large models faster and more cost-effective. It was trained on 45 trillion tokens of text, image, audio, and video, and can reason across all four. However, its current outputs are limited to text. The model is designed to provide calibrated answers, even flagging uncertainty, and lets users adjust "thinking effort" for speed. The company states Inkling uses a third fewer tokens than Nvidia’s Nemotron 3 Ultra for the same coding performance on one benchmark. Thinking Machines isn't claiming Inkling is the strongest model overall. Instead, it focuses on well-rounded performance and customizability, positioning it as a starting point for organizations to fine-tune using their Tinker platform. This approach challenges the idea of one-size-fits-all AI models, suggesting customized AI will perform better. This matters because it offers a different path for how organizations can develop and use AI tailored to their specific needs.