Thinking Machines Lab, under the leadership of former OpenAI CTO Mira Murati, has unveiled Inkling, its first foundation model fully trained from scratch and available with open weights. Inkling is positioned as an accessible, fine-tunable AI model that aims to bridge gaps in Western open-source AI technology.
- Inkling offers full open weights for developer customization and fine-tuning.
- Model features 975 billion parameters with efficient token usage for cost control.
- A Western open-source AI alternative amid dominance of Chinese models.
What happened
Thinking Machines Lab, founded by Mira Murati, has released Inkling, its first fully self-trained foundation model with open access to its weights. This model is designed to be downloaded, customized, and fine-tuned by developers and enterprises, challenging the prevailing Chinese dominance in open-weight AI models.
Inkling is a 975 billion parameter mixture-of-experts model trained on roughly 45 trillion tokens covering text, image, audio, and video modalities. It produces outputs exclusively in text form but supports complex data types such as code and structured results. The company highlights its model’s efficient use of parameters—only a fraction is active per prompt—to balance speed and operational cost.
Why it matters
The launch of Inkling fills a notable gap in the Western open-source AI landscape, which has been largely overshadowed by Chinese companies offering fully open-weight models. This release comes at a time when major players like Meta are shifting toward more proprietary AI frameworks, limiting customization options for external developers.
By providing an open-weights model with flexible fine-tuning capabilities and uncertainty flagging to reduce hallucinations, Thinking Machines is promoting accessibility, transparency, and control for organizations. This approach appeals to enterprises aiming to build tailored AI solutions without bearing high costs or dependency on external APIs.
What to watch next
Market observers will be keen to see how Western companies adopt Inkling compared to entrenched Chinese models and proprietary systems. Early collaborations, such as the work with Bridgewater Associates to create a specialized, cost-efficient financial reasoning model, hint at Inkling’s potential in vertical-specific AI applications.
Further development of features like Thinking Machines’ Tinker training API and enhancements in multimodal reasoning could significantly influence the open-source AI ecosystem. The model’s ability to balance customization, performance, and operating costs will be crucial to its broader acceptance and impact.