Thinking Machines Lab, an AI startup founded by Mira Murati, former Chief Technology Officer of OpenAI, officially launched its first self-developed open-source artificial intelligence model Inkling on July 15, 2026. This move marks the company's first public showcase after a year and a half of infrastructure development, aiming to directly challenge the "one-size-fits-all" closed-source AI business model dominated by major tech giants by providing highly customizable open-weight models.

Inkling uses a mixture of experts (MoE) architecture, with a total of 975 billion parameters, but only activates about 41 billion parameters for each task, effectively balancing ultra-large scale and operational efficiency.

The model is pre-trained natively on 45 trillion text, image, audio, and video tokens, and currently supports output in text, code, and structured data. Unlike general models that pursue "versatility," Inkling allows users to adjust the "thinking intensity" according to business needs to balance speed and accuracy. In code benchmark tests, it achieves the same performance with only one-third the token consumption of NVIDIA Nemotron3Ultra.

As the core of its ecosystem, Thinking Machines positions Inkling as the starting point for enterprise-level fine-tuning, paired with its model customization platform Tinker, monetizing through revenue-sharing fine-tuning and managed services. This strategy, which avoids the path of general chatbots and focuses on enterprise private customization, aligns with the current industry trend of preventing proprietary data leaks and pursuing cost reduction and efficiency improvement, and also opens up new pathways for the implementation of open-source AI in enterprise production environments.