Thinking Machines Lab releases first open model as proof that targeted AI works better

Thinking Machines Lab has published its first open model: Inkling. The San Francisco-based company, founded in February 2025, presents the model as concrete evidence of its central thesis that AI works better when it is specifically tailored to a task or context, rather than serving as a one-size-fits-all solution.

This is the first time the company has released anything tangible. Thinking Machines Lab spent eighteen months building its AI infrastructure largely without public visibility. Inkling marks the transition from a virtually unknown research organisation to a player with a public track record.

The company is structured as a public benefit corporation, a US legal form in which societal goals are legally enshrined alongside commercial interests. At the time of its announcement in February 2025, it had around thirty researchers and engineers. The company now counts between 130 and 170 employees, according to various sources.

What Inkling is and what it aims to do

Inkling is the first model from Thinking Machines Lab to be made publicly available. The name refers to the idea that AI does not need to know or do everything, but can instead be deployed in a focused way within a specific domain or use case. The company positions this as a counterweight to the dominant trend of ever-larger, more broadly trained models that use the same foundation for every application.

Concrete technical specifications for Inkling, such as the number of parameters, training data, or benchmark results, could not be fully verified based on available source material. Thinking Machines Lab has described the launch as a first public proof point, suggesting that further models or extensions are in the pipeline.

Releasing an open model is also a strategic choice. Open models can be inspected, modified, and deployed by third parties, which lowers the barrier for researchers and companies to work with them. At the same time, it makes Thinking Machines Lab's approach more transparent at a moment when the company still needs to prove itself in a crowded field.

Rapid growth and acquisition

Thinking Machines Lab grew in a relatively short period from a small research group to a company with over a hundred employees. In April 2026 it acquired Workshop Labs, a step consistent with a company looking to expand its capabilities quickly. Further details about the acquisition, such as the financial terms or exactly what Workshop Labs brought to the table, have not been made public.

No verified figures are available regarding Thinking Machines Lab's funding. The company has deliberately kept a low profile, making it difficult to assess its capital position or investor base. The growth in headcount points to substantial resources, but specific investment amounts have not been confirmed.

A broader debate about how AI is built

Thinking Machines Lab's position fits into a broader debate that has been ongoing in the AI sector for some time. On one side stand the large frontier models from companies such as OpenAI, Google, and Anthropic, trained on vast amounts of data and intended as broadly deployable systems. On the other side, there is growing interest in smaller, specialised models that are more efficient for specific applications and require less computing power during inference.

Thinking Machines Lab explicitly chooses the latter direction. Whether that choice pays off commercially depends on whether customers and developers actually need that specialisation and are willing to adopt a different foundation than the well-known large models.

With the publication of Inkling, the company gives the market something to react to for the first time. For European and Dutch AI companies and investors monitoring international developments, Thinking Machines Lab is an example of how a young company attempts to translate a substantive position on AI architecture into a recognisable market stance. The question of whether specialised, open models can become a genuine alternative to the broad frontier approach is also relevant for European initiatives working on sovereign or sector-specific AI applications.