Thinking Machines Lab, the artificial intelligence startup founded by former OpenAI Chief Technology Officer Mira Murati, has launched its first open weight foundation model, marking its formal entry into the increasingly competitive market for enterprise AI models.
Named Inkling, the multimodal model is being positioned as an alternative for developers and enterprises seeking greater flexibility than proprietary AI systems. The release comes as demand for open weight models continues to grow, with organisations looking for AI systems they can download, customise and deploy within their own infrastructure while retaining greater control over data and fine tuning.
Thinking Machines said Inkling has been trained from scratch using a mixture of experts architecture comprising 975 billion total parameters with 41 billion active parameters. The model supports a context window of up to one million tokens and has been pretrained on 45 trillion tokens spanning text, images, audio and video, enabling multimodal reasoning across multiple data formats.
Alongside Inkling, the company also previewed Inkling-Small, a lighter version designed to deliver lower latency and reduced deployment costs while maintaining strong performance for enterprise applications. According to the company, both models are intended to balance reasoning capability with computational efficiency rather than competing solely on benchmark scores.
Unlike closed source models where only APIs are accessible, Inkling is being released with open weights, allowing developers to download, adapt and fine tune the model for domain specific applications. Thinking Machines said the approach aligns with its broader vision of building AI systems that users can customise instead of relying entirely on centrally managed foundation models.
The launch also strengthens competition in the open model ecosystem, where offerings from NVIDIA's Nemotron family, Google's Gemma models, Meta's Llama series and several Chinese AI companies have accelerated enterprise adoption of customisable AI. Industry observers view Inkling as another step towards expanding enterprise choice, particularly for organisations prioritising data governance and private deployments.
Thinking Machines said Inkling is designed to reason natively across text, images and audio while offering controllable "thinking effort" that allows developers to balance response quality with computational cost. The company added that enterprises can further customise the model through its AI platform, Tinker, which supports fine tuning and deployment without requiring organisations to build complex AI infrastructure from scratch.
The release follows Thinking Machines' rapid rise since its launch in 2025. Founded by Murati after leaving OpenAI, the startup has attracted significant investor interest, raising one of the largest seed funding rounds in the AI sector. Earlier this year, the company also announced a long term infrastructure partnership with NVIDIA to build large scale AI training systems powered by the chipmaker's next generation Vera Rubin platform.
Industry analysts note that the introduction of Inkling reflects a broader shift in enterprise AI adoption. Rather than relying exclusively on proprietary models, organisations are increasingly evaluating open weight alternatives that provide greater transparency, customisation and deployment flexibility. This trend has intensified as businesses seek AI systems tailored to industry specific workflows while maintaining tighter control over sensitive data.
With Inkling now available, Thinking Machines joins a growing group of AI companies competing to shape the next phase of enterprise AI, where openness, efficiency and adaptability are becoming as important as raw model performance.
Related Articles
