Thinking Machines Lab, the AI startup led by Mira Murati, has released Inkling, an open-weight mixture-of-experts foundation model with 975 billion total parameters and 41 billion active parameters per query. The company says the model weights are available on Hugging Face, while fine-tuning is supported through its...
Thinking Machines Lab released Inkling, its first open-weight foundation model, according to the company’s announcement and the model card published on Hugging Face.
Thinking Machines Lab describes Inkling as its first open-weights model and says it is available through Hugging Face, with fine-tuning support offered through the company’s Tinker platform. Axios reported that the July 15, 2026 release is also the company’s first AI model and first foundation-model release.
The Hugging Face model card published by Thinking Machines identifies Inkling as a mixture-of-experts model with 975 billion total parameters and 41 billion active parameters. In a mixture-of-experts architecture, only part of the model is used for a given request, which can reduce the amount of computation needed at inference compared with activating the full parameter count.
Thinking Machines’ announcement frames the release around open-weight access, meaning developers can download the model weights rather than only access the system through a hosted API. The company says Inkling is available on Hugging Face, and Axios also reported that the weights are being released there.
According to Thinking Machines Lab, Inkling can be fine-tuned through Tinker, the company’s platform for customizing models. Axios reported the same availability path: weights on Hugging Face and fine-tuning through Tinker.
Open-weight availability can matter for organizations that want more control over deployment, evaluation, customization, or data handling. However, the sources provided do not establish real-world enterprise adoption, benchmark leadership, or cost savings in production, so those claims should be treated as unverified until independent evaluations are available.
Zonebourse, citing the launch in French-language coverage, described Inkling as an open model positioned as a Western alternative to Chinese open models. That framing reflects the broader competitive environment around open-weight AI systems, where companies and research groups in multiple regions are releasing large models for developers to inspect, adapt, and run in their own environments.
The available sources do not provide independent benchmark comparisons against Chinese open models or other leading systems. For now, the concrete facts are the release itself, the model’s architecture and parameter counts as stated on Hugging Face, and the distribution and fine-tuning options described by Thinking Machines and reported by Axios.
Inkling marks a notable first release for Thinking Machines Lab because it puts the company into the open-weight foundation-model market rather than limiting access to a closed hosted service. The release gives researchers and developers a new large-scale mixture-of-experts model to evaluate directly, while also giving Thinking Machines a public foundation for future tooling and model development.
The next important test will be independent assessment: how Inkling performs across reasoning, coding, multilingual tasks, safety evaluations, and deployment costs compared with other open-weight models. Until those results are available, the launch is best understood as a significant new model release from a high-profile AI startup, not yet as proof of technical leadership.
Thinking Machines Lab released Inkling, its first open weight foundation model, according to the company’s announcement and the model card published on Hugging Face.
Axios reported that the July 15, 2026 release is also the company’s first AI model and first foundation model release.
The Hugging Face model card published by Thinking Machines identifies Inkling as a mixture of experts model with 975 billion total parameters and 41 billion active parameters.
Continue reading
