Thinking Machines Lab launches Inkling, a 975-billion-parameter open-weight AI model
The release of Inkling marks Thinking Machines Lab’s first major product, positioning the firm to compete directly with OpenAI and Anthropic in the open-source market.
Thinking Machines Lab, an artificial intelligence startup established in February 2025 by former OpenAI executives, has released its inaugural model, Inkling. The 975-billion-parameter open-weight system is trained to process audio, video, and text inputs, marking a significant step in the company’s bid to challenge industry leaders such as OpenAI and Anthropic. The release supports the firm’s stated vision for a decentralised AI ecosystem, arguing that technology should not be controlled by a small number of corporations.
Founded by Mira Murati, John Schulman, and Lilian Weng, the startup secured the largest seed funding round in history, valuing the company at $12 billion. Prior to Inkling, the firm had released Tinker, a tool for fine-tuning models, and published various machine-learning research papers. The new model is designed to be downloaded and modified by researchers and startups, contrasting with the closed-model approaches that dominate much of the current market.
Inkling was trained from scratch to understand audio and video input in addition to text. While the company acknowledges that the model is not the best on popular benchmarks, it states that Inkling offers performance levels similar to the top open-weight models currently available from China. The system is capable of advanced reasoning and coding, though it requires a cluster of specialised chips to run due to its size.
During the training process, researchers observed that Inkling initially decided to remove natural language explanations for its reasoning to improve efficiency, viewing grammar as overhead. The company subsequently reinstated natural language reasoning to enhance explainability. The firm also used Inkling to fine-tune and improve itself, illustrating the growing trend of AI models being used to build other AI models.
The launch of an open-weight model aligns with the company’s strategy to offer a cheaper and more flexible alternative to closed systems. Open-source models have gained popularity due to lower running costs and greater ease of modification for specific tasks. This move places Thinking Machines Lab in direct competition with defector-led companies like Anthropic, which recently filed for an IPO potentially valuing it at over $1 trillion.
