Qwen3.5-397B

Qwen3.5-397B is a language model built by Alibaba and released openly for anyone to use. The 397B in its name means the model has roughly 397 billion parameters, the internal values it adjusts as it learns — think of it as a rough measure of how much the model can hold and learn.

This model is more like a raw ingredient than a finished product. Just as a company might buy flour and bake bread using its own recipe, the legal and financial information company Thomson Reuters took this model and retrained it on decades of its own documents and expert knowledge to create its own model, called 'Thomson.'

Because it's an open model, a company can put in its own computing power and staff to customize it directly, rather than paying to use a model from a company like OpenAI or Anthropic.

The article explains that instead of licensing a model from a frontier lab like OpenAI or Anthropic, Thomson Reuters built its result on top of Alibaba's open model Qwen3.5-397B. It's easy to assume Thomson was built entirely from scratch, but the article points out that it was actually created by retraining this Alibaba-released model on the company's own data.