Mira Murati’s Thinking Machines drops Inkling, an open-weights model anyone can access

What changed

Thinking Machines Lab Inc., led by Mira Murati, released Inkling, its first foundation model with fully open weights. Unlike many AI models that keep their parameters closed or partially accessible, Inkling’s open weights let developers fine-tune the entire model from scratch. This is the first fully trained model released by Thinking Machines after a period focused mostly on research and experimentation.

Why builders should care

Open-weights models reduce barriers around customization. Instead of adapting pre-tuned closed models within strict vendor limits, developers can now adjust Inkling’s entire architecture to fit specific needs. This flexibility cuts costs for builders who otherwise pay for expensive fine-tuning by third parties or rely on restrictive APIs. It also shifts some control back to operators by lowering vendor lock-in risks and opening paths to deeper domain adaptation.

The practical takeaway

If a development team has enough compute and expertise, Inkling allows more precise tailoring for vertical applications, new languages, or unique workflows. This model release signals a push for openly accessible AI foundations that developers can own and evolve directly. It pressures proprietary models by offering an alternative with no paywalls on the core weights. That could speed feature innovation and differentiation among startups and enterprises alike.

What to watch next

Follow how the developer community adopts Inkling in real projects, especially whether open weights translate to meaningful model improvements or accelerated time to market. Track whether other players in the foundation model space respond by loosening restrictions or opening weights themselves. Also watch for how Thinking Machines evolves this model and whether more specialized or scaled-up versions emerge that broaden the ecosystem further.

AI Quick Briefs Editorial Desk