Nvidia and Reflection Release Free AI Models Built for Business

Nvidia and Reflection each released AI models built for business work, and companies can run them on their own servers. For most of the year, Chinese labs led this part of the AI race. This week, U.S. companies put up their own entries, and two Chinese labs, DeepSeek and MiniMax, cut prices again.

What changed

These are called open-weight models. The company publishes the model itself, so a business can run it in-house instead of sending data to someone else's service. Ramp and EY are already adjusting Nvidia's model to read financial documents. Reflection launched with more than 40 partners, including AWS and Microsoft.

DeepSeek and MiniMax shipped updates that finish tasks faster for less money. MiniMax claims a 70% cost reduction. Inception says its Mercury 3 model answers four to eight times faster than rivals. Perplexity's new models make quick yes-or-no calls, such as routing requests or screening them.

Who it affects

Banks and insurers that could not use Chinese models for security reasons now have a wider choice of what to run themselves. Any business that handles sensitive data may find the same.

What is not known yet

The material does not say how the new U.S. models compare with Chinese ones on quality or price. It also does not give the names of the specific models, their license terms, or how they will be priced. The 70% savings and the speed gains are the companies' own claims.