China’s DeepSeek in New Challenge: Innovative Model Competes with GPT-5

The Chinese artificial intelligence company “DeepSeek” has announced the release of two new versions of its intelligent model, after launching the trial version a few weeks ago. The first version, DeepSeek-V3.2, is a direct update to the previous model, and the company claims that it rivals the performance of GPT-5 in many reasoning tests, according to a “Bloomberg” report.

This model integrates the simulation of human thinking and the ability to use various digital tools, such as search engines, calculators, and code execution programs. The company confirmed via its account on the “X” platform that it is the first model to directly link thinking to the mechanism of using tools.

As for the second version, DeepSeek-V3.2-Speciale, it focuses specifically on computational and inferential capabilities, and the company claims that it matches the performance of Gemini-3 Pro from “Google.” This model has achieved outstanding results in international tests, such as the Mathematics Olympiad and the Informatics Olympiad.

The company indicated that it has developed an innovative method for training artificial intelligence agents, as part of its continuous endeavor to make the models faster and more efficient. “DeepSeek” continues to strengthen its position in the Chinese artificial intelligence race, after the model that sparked widespread interest at the beginning of the year, and its release last week of the DeepSeekMath-V2 model specialized in proving mathematical theorems.