This week, Thomson Reuters presented a different path. The company announced Thomson, its first proprietary large language model, built on an open-source foundation and specialised with its legal, tax, regulatory and news archives. It says it invested $40 million across talent and compute and retains full control of the system. Its performance and cost claims still rely largely on company evaluations, while broader external validation continues.
From a K2Reframe perspective, the decisive question is not how large the model is. It is what responsibility the model carries, which sources inform its decisions and how its output can be verified.
For the C-suite, the question is shifting from “Did we buy the most powerful model?” to “Which business, risk and accountability frame did we put this intelligence inside?”
Do you think enterprise AI competition is moving from scale toward specialisation and verifiability?