Thomson Reuters develops own AI model using Alibaba's open-source technology
The company aims to reduce reliance on Claude due to high costs.
1Birth of Thomson-1
Thomson Reuters developed its AI model Thomson-1 by 'realigning' Alibaba's Qwen3.5 model. CTO Joel Hron explained that the system, called Snowdon, will focus on areas of company expertise like document review. The open-source model stands out for being freely downloadable and customizable. This move aims to reduce the company's dependence on expensive AI solutions like Anthropic's Claude.
2Cost-driven transformation
High usage costs of chatbots are pushing major companies to seek alternatives. By developing its own model, Thomson Reuters plans to eliminate ongoing payments to external AI companies. Hron likened this to 'buying a house instead of renting'. The CoCounsel legal assistant, part of an expanded partnership with Anthropic in May, will continue to primarily use Claude for now.
3Shift to Chinese models
Western tech companies are increasingly adopting open-source AI models from China. Many startups are building products using these models. However, this trend brings controversies. Anthropic claims Chinese labs are 'distilling' AI outputs to train new chatbots. Figures like Senator Tom Cotton warn about security threats and backdoor risks.
4Ethical and safety measures
Thomson Reuters states it made the Qwen model 'ethically and politically unbiased' through months of work with Imperial College. Hron emphasizes the model underwent extensive testing for safe use. The company believes building its own IP will create long-term value. This approach represents a significant example of the self-sufficiency trend in AI.
5Industry impact
This move by legal and data giant Thomson Reuters could mark an industry turning point. Experts predict companies in high-license-cost fields may develop similar strategies. Internalizing AI infrastructure could provide competitive advantages in specialized sectors. This might pave the way for customized corporate AI solutions.
6Tech geopolitics
The rise of Chinese AI models creates new competition in tech. Western companies adopting Chinese models raises national security concerns. Some US politicians oppose companies like Airbnb using Chinese models. These developments may lead to new regulations in tech supply chains.
7Long-term strategy
Thomson Reuters aims to develop Thomson-1 to power more CoCounsel features over time. While expanding its IP portfolio, the company seeks AI independence. This strategy could offer significant advantages in specialized fields like legal tech. Internalizing AI infrastructure may provide critical competitive edges.
8Domino effect in industry
Thomson Reuters' move could create a domino effect, especially in sectors with high licensing costs. Experts predict similar in-house AI projects may increase in finance and healthcare. This trend might challenge the dominance of major AI players. Companies developing specialized models could establish new industry standards.