rewrite this title DeepSeek Announces New AI Model That Slashes Costs By Half!

DeepSeek has announced its new, currently experimental artificial intelligence model, the “V3.2-exp.” This model appears capable of halving API access costs thanks to a special technology it possesses. Let’s see if DeepSeek’s claim will become a reality… A significant move has come from the Chinese AI startup DeepSeek. The company announced a new artificial intelligence model named “V3.2-exp.” This AI model, which is still in the experimental stage, features a very interesting capability that distinguishes it from its competitors. In fact, thanks to a specialized technology, this AI model can reduce API costs by up to 50 percent. Let’s delve into the details.

DeepSeek’s new AI model utilizes a technology called “DeepSeek Sparse Attention.” This special technology uses a module named “lightning indexer” to prioritize the context based on the command entered by the user. As a result, the artificial intelligence only generates a context window relevant to the command. In the second stage, a “fine-grained token selection system” takes over, subcategorizing the initial context window once again. This reduces the server load, thus lowering the API cost.

Will Provide an Advantage for Long-Term Tasks!

According to DeepSeek’s statement, the DeepSeek Sparse Attention technology will provide an advantage to those who use AI models for long-term tasks. As we mentioned at the beginning of our news, costs will be cut by half. Of course, it is important to remember that these are currently just DeepSeek’s claims. Independent tests will reveal whether the V3.2-exp model is truly inexpensive. One of the most significant problems with artificial intelligence technologies is the rising prices of API access. If DeepSeek delivers on its claim, it appears it could open the door to new competition in the sector and potentially put rivals like OpenAI, Microsoft, and Google in a tight spot. We’ll have to wait and see how this all turns out…

You Might Also Like;

Copy URL
URL Copied