TL;DR: Thomson Reuters released 'Thomson,' a specialized LLM built on Qwen3.5-397B and fine-tuned with 175 years of proprietary legal and financial data, claiming performance comparable to top models.
Summary: Thomson Reuters has launched 'Thomson,' a new large language model. It leverages Qwen3.5-397B as its base and undergoes continuous training on the company's extensive 175-year archive of legal, tax, accounting, and news data. The company reports that Thomson performs comparably to Claude Opus 4.8 and surpasses other leading models on their internal benchmarks.
Why it matters: This demonstrates the growing trend of large enterprises developing highly specialized LLMs using proprietary data for domain-specific applications. AI builders should consider the potential for niche, high-value datasets to create competitive models, and watch for opportunities to integrate with or build on such specialized platforms.
Source: x_com