Thomson Reuters unveils its own frontier‑class AI, trained on decades of legal and financial data

On August 24, information‑services giant Thomson Reuters announced “Thomson,” its first in‑house large language model positioned as a frontier‑class system built for professionals. The model is trained on the company’s deep archive of legal, tax and financial content, aiming to power tasks like drafting, research and summarization in tightly regulated domains.

Rather than spending billions of dollars on compute, as many frontier labs have done, Thomson Reuters says it invested about $40 million and built on an open‑source foundation to tailor the model to high‑value professional workloads. The company also stresses that inference costs are significantly lower than typical frontier models, making it easier to embed the AI directly into customers’ existing tools and workflows.

The launch reflects a broader shift toward “sovereign” and domain‑specific AI, as enterprises worry about where models run, how they are trained, and how client data is protected. Thomson Reuters plans to weave the new model into its legal research and tax platforms, enabling features like case‑law analysis and contract review with stronger privacy controls than generic cloud chatbots. How far these industry‑tuned models can challenge general‑purpose AIs in day‑to‑day professional use is likely to be a key theme in the next phase of AI adoption.