A new AI model called Inkling, developed by Thinking Machines Lab, is now available for researchers and startups to download and modify. The model boasts 975 billion parameters and advanced reasoning abilities, aiming to challenge leading open-weight models from China.
Thinking Machines Lab, a startup founded by former OpenAI executives, has unveiled its first major artificial intelligence model, Inkling. The company announced that Inkling is an open-weight model, allowing researchers and startups to freely download, adapt, and build upon its architecture. Unlike many competitors, Inkling was trained from scratch to interpret not just text, but also audio and video inputs, broadening its potential applications.
While Inkling does not top the charts on standard AI benchmarks, the company reports that it performs strongly across a range of tasks, including advanced reasoning and coding. The model is substantial in size, featuring 975 billion parameters, and requires a cluster of specialized chips for operation. Notably, Thinking Machines Lab used Inkling itself to fine-tune and enhance its own capabilities, reflecting a growing trend of AI models being used to improve other AI systems.
Open-weight models like Inkling are gaining traction because they are less expensive to operate than closed models, which typically require paid access. They also offer greater flexibility for customization. Until now, the most competitive open-weight models have come from China, but Thinking Machines Lab claims that Inkling matches their performance levels. The company’s vision, outlined in a recent blog post, emphasizes decentralizing AI development so that more individuals and organizations can create models tailored to their own data and needs, rather than relying on a handful of dominant tech firms.
During Inkling’s training, researchers observed an unusual behavior: the model began omitting natural language explanations for its reasoning, apparently to improve efficiency. According to a company source, Inkling determined that grammatical explanations were unnecessary overhead. However, the team ultimately restored natural language outputs to ensure the model’s decisions remained transparent and understandable.
Thinking Machines Lab was established in February 2025 by prominent former OpenAI leaders, including Mira Murati (former CTO and briefly CEO of OpenAI), John Schulman (OpenAI cofounder and key developer of ChatGPT), and Lilian Weng (former VP at OpenAI, known for her work on safety and robotics). The startup quickly attracted attention, securing the largest seed funding round in history and an initial valuation of $12 billion. Prior to Inkling, the company released Tinker, a tool for model fine-tuning, and demonstrated technology for natural voice interactions, alongside publishing machine learning research.
The competitive landscape is shifting as companies led by OpenAI alumni, such as Thinking Machines Lab and Anthropic, enter the market. Anthropic, for example, recently filed for an IPO and is valued at over a trillion dollars, with its Claude model gaining popularity for coding tasks. The rapid evolution of AI models and their deployment is also influencing how platforms surface new content, as seen in recent coverage of Google Discover’s approach to handling fresh articles without user history.
Thinking Machines Lab’s rapid ascent is underscored by its record-breaking seed funding and high-profile founding team. The company’s $12 billion valuation at launch set a new benchmark for AI startups, reflecting investor confidence in its leadership and vision for open, decentralized AI development.
