A Shift in the AI Industry: 'Thomson', a Proprietary Model for Legal Professionals, Has Arrived

AI Summary

Thomson Reuters has emerged as a new powerhouse in the professional AI market with the launch of 'Thomson,' its proprietary AI model trained on decades of accumulated professional data.

Imagine you need to review complex legal documents, and you have an AI assistant that perfectly understands the context of your work. The artificial intelligence (AI) we have been using until now has been like a “knowledgeable college student” who has broadly learned all the information in the world. However, an “AI veteran expert” has now emerged that shows sharper insight than any college student in specific fields. This is the story of ‘Thomson,’ a proprietary AI model announced by legal information service giant Thomson Reuters.

Why is this important?

Until now, the AI industry’s frontier models have been dominated by large technology companies like Google or OpenAI, pouring billions of dollars into development. However, rather than recklessly investing a fortune, Thomson Reuters chose the path of building its own model with its own expertise, investing approximately $40 million over two years [Source 2, Source 4].

This signifies that companies have entered an era where they no longer rely solely on renting external general-purpose models but are taking ownership of AI by utilizing their own ‘data assets’ [Source 4, Source 11]. In fields where security and accuracy are paramount, such as law or finance, the value of a specialized model armed with trusted data is inevitably higher than any general-purpose model.

Simplified: Adding ‘Advanced Major’ to ‘Basic Education’

To put it in perspective: if general-purpose AI models are smart students who have finished primary and secondary education, Thomson is like putting that student through an advanced ‘legal expert’ major program.

Rather than starting from scratch, Thomson Reuters used Alibaba’s proven open-source model ‘Qwen’ as its foundation [Source 4, Source 10]. They then intensively trained it on the legal, tax, and regulatory data that Thomson Reuters has accumulated over several decades [Source 10].

Simply put, they took an AI that already possessed basic general knowledge and fed it entire libraries of law books and the latest legal precedents, turning it into a legal expert. As a result, Thomson exhibits a level of accuracy in professional domains that other models find difficult to imitate. Benchmark tests actually showed that in the legal field, Thomson stands shoulder-to-shoulder with top-tier models like Claude Opus and, in certain metrics, even outperformed GPT-5 [Source 7].

Current Situation: Expert Verification Phase

Currently, Thomson Reuters is using this model internally first, while also sharing it with groups of legal and AI experts to undergo a direct evaluation process [Source 1, Source 3].

Furthermore, the Thomson model is scheduled to be applied as the default model for one of Thomson Reuters’ services, the ‘Tabular Analysis’ feature. Of course, the company guarantees users the choice to select other models if they prefer [Source 2]. This strategy is not about unconditional monopoly, but about helping users make the optimal decision between efficiency and expertise.

What lies ahead?

The arrival of Thomson is changing the landscape of the AI industry. It is no longer the case that top-tier AI is the exclusive property of massive technology companies [Source 5].

Moving forward, Thomson Reuters plans to continuously develop and verify this model with external partners [Source 1]. What we should pay attention to is the possibility of this model expanding beyond law into wider fields such as tax and regulatory compliance. Much like the ambition of Thomson Reuters, which stated, “We are not riding the AI wave; we are designing the future ourselves,” the work methods of specific professions are now ready to be fundamentally transformed through AI [Source 13].

MindTickleBytes’ AI Reporter View

This case proves that AI does not necessarily need to be large and general-purpose. Rather, knowledge deeply rooted in a specific field is the most powerful weapon in the AI era. Thomson Reuters’ move will remain a textbook example of how companies with data can reclaim leadership in the AI era.

References

Share this article:

Test Your Understanding

Q1. What did Thomson Reuters base its 'Thomson' model on during development?

Google's Gemini
Alibaba's Qwen
OpenAI's GPT-5

Thomson Reuters developed 'Thomson' by using Alibaba's open-source model, Qwen, as a foundation and training it on its own professional data.

Q2. What is the most significant feature that distinguishes the Thomson model from other general-purpose models?

It can process all types of conversations
It was trained on decades of accumulated legal and regulatory expertise
It costs nothing to develop

Thomson has been optimized for legal and regulatory tasks by training on Thomson Reuters' proprietary and professional content, such as Westlaw.

Q3. Approximately how much did Thomson Reuters invest in this AI model development?

$4 million
$40 million
$4 billion

Thomson Reuters invested approximately $40 million over two years for this project.