The Chinese artificial intelligence startup DeepSeek has launched preliminary versions of the V4 model, featuring significant improvements, in the context of competition with AI from the USA.

DeepSeek launched preliminary versions of its model V4 on Friday, eagerly awaited by users who want to compare it with American competitors like ChatGPT and Claude. DeepSeek has been accused by OpenAI and Anthropic of developing technology based on their models. The new V4 model includes 'pro' and 'flash' versions, with significant improvements in knowledge and reasoning capabilities. Analyzing the performances, DeepSeek claims that V4 Pro Max surpasses GPT-5.2, but remains behind GPT-5.4. Additionally, V4 has a context window of 1 million tokens, a considerable improvement over V3. Despite these advancements, some analysts remain skeptical, emphasizing that independent evaluations are necessary to confirm the model's capabilities. Furthermore, DeepSeek has been accused of unethical practices in the development of its technology, and the Chinese embassy in the U.S. has rejected these accusations.