Thomson Reuters’ New Model Could Inspire Other SaaS Vendors

Information services vendor Thomson Reuters launched its own proprietary model on Monday, which it said it trained at less than half the cost of other frontier AI models. The release shows what is possible for SaaS vendors like Thomson Reuters that are looking to capitalize on the AI market.Developed in-house, Thomson was built on the vendor’s proprietary content -- with a focus on legal information -- technology and domain expertise. The model starts with a base model called Snowdon, developed by the FAIR Lab at Imperial College London.Thomson Reuters, based in Toronto, frames its model, Thomson, as comparable to the strongest frontier models on the market, including Claude Opus 4.8, GPT 5.5 and Gemini 3.1 Pro. The new model comes after SaaS vendors in the legal information services sector were shaken following Anthropic's release of Claude Cowork plugins in February.In addition to law, Thomson Reuters is aiming the new model at professionals in fields such as accounting and other compliance-related areas.Related:Waymo Develops Its Own Chip for Self-DrivingThomson Reuters said the model's origin also dates to its acquisition of Safe Sign Technologies in 2024. Safe Sign was a U.K. AI startup that developed legal-specific large language models (LLMs). The startup’s staff became Thomson's foundational research team.The Model Could Spur Others“This could be an inspirational model for other institutions that are also sitting on top of massive reserves of intellectual property and capital,” said Michael G Bennett, associate vice chancellor for data science and AI strategy at the University of Illinois Chicago.He added that for vendors, especially for organizations with data that has not been absorbed by frontier model makers, creating their own model for commercial use may make sense.“This will be an eye-opener for any number of sectors that have been thinking about how to embrace the technology simultaneously and at the same time make use of whatever intellectual expertise, resources they have internally to build not only a model that’s really powerful and useful, but one that is also distinguishable from a frontier model,” Bennett continued.An Inexpensive EndeavorThe route Thomson Reuters used to create its model is also intriguing, Bennett said. By using an open-weight model as the basis for Thomson, Thomson Reuters only had to invest $40 million to train it, much less than the cost of training an advanced LLM. The typical cost can exceed $100 million.“Their expense was probably on the order of one or two magnitudes less than what it would take to build a model from the ground up,” Bennett said.Related:Ode With Anthropic Makes First Acquisition to Expand Enterprise AIIt is unclear how Thomson’s model will perform in the market, but the LLM ought to perform well technically, Bennett said. For companies in knowledge industries, the ability to train domain-specific models without incurring the typical expenses of model development and using an inexpensive model could be attractive, he added.While Thomson Reuters overcame organizational and technical hurdles in building the LLM, there are still some challenges for the vendor, including convincing its customers that the model is comparable to other frontier models, Bennett said.Enterprises subscribing to Thomson will also need to discover and compare for themselves whether an LLM like Thomson, with domain expertise, is comparable to those from frontier AI labs.About the AuthorNews Writer, AI BusinessEsther Shittu has covered AI technologies and industry trends since 2021. As co-host of the Targeting AI podcast, she talks with experts, thought leaders and practitioners exploring critical AI developments. Before AI Business, she wrote for SearchEnterpriseAI, the New York Daily News, Bklyner and the Brooklyn Daily Eagle. When she's not diving deep into the world of AI, she spends her time on passion projects and raising her three daughters.