Lawyers, start your laptops. A new custom AI tool has arrived that (hopefully) won’t hallucinate and make up legal cases.
Driving the news: Toronto-based software provider Thomson Reuters launched its own custom AI model yesterday that will cater directly to legal and accounting customers. The new model, which bears the uber-creative name Thomson, was built in-house with Alibaba’s open-source model Qwen for just US$40 million.
The SaaSpocalypse selloff has sent Thomson Reuters stock cratering nearly 40% over the past year. This model is part of a broader push by it and other software companies to win back business customers from the likes of Anthropic and OpenAI.
Why it matters: For the past couple of years, it seemed as though every business was willing to pay any price for access to the newest, most advanced AI model (mostly out of fear of falling behind competitors). Now, staring down mounting token costs, more companies are realizing that cheaper, customized AI models can be just as effective.
Airbnb, DoorDash, and Siemens have all recently adopted and customized Chinese AI models — DoorDash even tiered its AI model usage based on the level of work.
It appears this trend is already hurting Anthropic. Two months after Fable’s release, the model makes up only 11% of total spending on Anthropic tools.
Our take: For many businesses, products like Fable are overkill; a boutique law firm doesn’t necessarily need to be able to vibe code. If cheaper, custom models can master specific use cases for businesses, frontier labs could see an exodus of enterprise customers.—LA
