Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released Inkling, a 975 billion parameter mixture-of-experts model trained on 45 trillion multi-modal tokens, designed for enterprise fine-tuning. The model was trained on Nvidia's GB300 NVL72 systems, with the company having secured a partnership for a gigawatt of Vera Rubin computing capacity. Inkling claims to use one-third the tokens of Nvidia's Nemotron 3 Ultra for the same coding performance and achieved 84.7% on financial reasoning tests in a joint project with Bridgewater Associates.
A new open-weight model from a frontier lab challenges the closed-model revenue paradigm by emphasizing efficiency and customer-led fine-tuning, shifting value to customization platforms.