Key Highlights

• Thinking Machines has unveiled Inkling, its first open-weight AI model.
• The model features 975 billion parameters but activates only 41 billion during inference.
• Inkling is designed for enterprises that want to customize AI models for their own needs.
• The launch positions Thinking Machines against closed AI offerings from OpenAI, Google, and Anthropic.

Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has introduced its first in-house AI model, Inkling. Unlike leading AI systems such as ChatGPT, Gemini, and Claude, Inkling is an open-weight model, allowing developers and enterprises to download, modify, and fine-tune it for their own applications.

The release marks the company’s first major product launch after spending more than a year building AI infrastructure behind the scenes. It also signals a strategic challenge to the closed-model approach that currently dominates the AI industry.

What is Inkling and how does it work?

Inkling is a mixture-of-experts (MoE) model with 975 billion total parameters. However, it activates only about 41 billion parameters for each task, making it more efficient than traditional large language models.

According to Thinking Machines, the model was trained on 45 trillion tokens spanning text, images, audio, and video. Although it can reason across all four data types, its current outputs are limited to text, including code, structured data, and formatted documents.

The company has also introduced features that let users adjust the model’s “thinking effort,” allowing them to trade deeper reasoning for faster responses depending on the task.

Another notable feature is calibrated reasoning. Instead of confidently generating uncertain answers, Inkling is designed to acknowledge uncertainty when it lacks confidence.

How is Inkling different from ChatGPT and Gemini?

Thinking Machines is taking a different path from OpenAI, Anthropic, and Google.

Rather than building a one-size-fits-all AI assistant, the company believes organizations should customize models using their own expertise. Inkling is positioned as a foundation model that enterprises can fine-tune through the company’s customization platform, Tinker.

This strategy reflects a growing debate in the AI industry over open versus closed models.

While proprietary platforms like ChatGPT, Claude, and Gemini prioritize centralized updates and managed services, Thinking Machines argues that organizations achieve better results when AI systems are trained on domain-specific knowledge.

The company points to a recent collaboration with hedge fund Bridgewater Associates, where an open-source model trained on proprietary financial knowledge reportedly outperformed leading closed AI models in financial reasoning while operating at a fraction of the cost. However, those findings have not been independently verified.

Why does this launch matter?

Inkling arrives as interest in open AI models continues to grow. Industry leaders, including Microsoft CEO Satya Nadella and Hugging Face CEO Clem Delangue, have recently argued that enterprises may increasingly favor customizable or private AI systems over proprietary models for production workloads.

Thinking Machines also claims Inkling reached the market in just nine months — considerably faster than the timelines reported by OpenAI and Anthropic for their first commercial products.

The company acknowledges that Inkling is not currently the strongest AI model available. Instead, it is positioning the model as a flexible starting point for organizations that want greater control over performance, costs, and data.

With Inkling, Thinking Machines is entering one of the AI industry’s biggest debates: whether the future belongs to centrally managed AI assistants or open models that businesses can tailor to their own expertise. That question could shape the next phase of enterprise AI adoption.