July 31, 2026
Objection! AI bragging in court
Thomson Reuters built its own AI model that now ranks among the best
Thomson Reuters says its homegrown AI is elite — and the comments instantly got snarky
TLDR: Thomson Reuters says its new AI model performs near the top of the field, especially on legal and professional tasks, showing big companies outside the usual AI giants may now be serious players. The comment section’s reaction was a skeptical eyebrow raise, with readers mocking the company for praising its own results.
Thomson Reuters just dropped a big flex: it says its new in-house AI model, Thomson, can hang with the biggest names in artificial intelligence and even beat some headline-grabbing rivals on legal work, coding, tax, accounting, journalism, and more. The company’s pitch is simple enough for non-experts: instead of building a chatbot that knows a little about everything, it built one trained on decades of its own trusted legal and business material, the kind professionals actually use when mistakes can be expensive.
But the community reaction? Instant side-eye. The standout mood was pure "show us, don’t tell us." The thread’s sharpest jab came from commenter blitzar, who dryly mocked the self-congratulatory tone with: they do say so themselves. That one-liner basically became the whole vibe. Readers weren’t exactly throwing confetti over a company grading its own homework, especially when the claims included beating famous AI names like GPT and Claude on selected tests.
That’s where the drama lives: is this a genuine underdog power move, or a corporate victory lap based on cherry-picked scorecards? The article insists the model was shaped by real experts and won’t train on customer data, which is the sort of promise that matters in law and finance. But in the comments, the biggest benchmark was trust. And right now, the crowd seems entertained, skeptical, and very ready with the memes.
Key Points
- •Thomson Reuters announced early benchmark results for its new AI model, Thomson, and said it performed competitively with leading frontier models.
- •The company says Thomson outperformed GPT-5.5, Claude Sonnet 5, and Gemini 3.1 Pro in the cited early benchmarking, while competing with Claude Opus 4.8.
- •Thomson Reuters says the model was built on an open-source foundation and further trained using proprietary content from Westlaw, Practical Law, Checkpoint, and Reuters.
- •The article says hundreds of subject matter experts evaluated outputs, identified failure modes, and validated reasoning aligned with legal professional workflows.
- •Thomson Reuters says Thomson is smaller and cheaper to train and operate than leading models, and that less than 10% of the company’s content has been used in training so far.