DeepSeek attacks the American giants with a new open-source AI that is achieving remarkable feats. / CheapToken News

Published on April 24, 2026 at 10:14

Commenter

DeepSeek has unveiled two new models: DeepSeek V4 Flash and DeepSeek V4 Pro. The latter would be capable of processing 1 million characters at once. This launch is part of the competition between China and the United States in the field of open-source AI.

DeepSeek announced « the era of a million-character context at a lower cost ». Two models were presented: DeepSeek V4 Pro and DeepSeek V4 Flash. The first, multimodal, would be capable of memorizing and understanding a million characters at once, while the second is less performant, but more economical.

A new model released in two versions

DeepSeek announced in a statement posted on the Chinese social media platform WeChat: « Today, the pre-version of our entirely new model series, DeepSeek V4, is officially launched and published as open source ». Specifically, this new model is available in two versions: DeepSeek V4 Pro and DeepSeek V4 Flash.

The model DeepSeek V4 Pro, multimodal, is capable of processing text, images, and videos. The company explains that « by comparison to the previous generation, the agent capabilities of DeepSeek V4 Pro have been significantly enhanced ». However, it's probably its text processing that impresses the most. The Chinese company explains that its model : « has an ultra-long context of a million characters ».

In other words, the model can memorize and understand a million characters at once. Thus, if given hundreds of pages, it will be able to assimilate the entire corpus in record time and answer any question about the content. DeepSeek claims to have made this process faster and cheaper. Its ability to process long texts would be « among the best in the world ». As for DeepSeek V4 Flash, it is a less powerful model but more « cost-effective ».

Open source AI: the war between China and the United States continues

The launch of DeepSeek V4 Pro takes place in a particular context. Early April, Google announced the launch of the Gemma 4 family, which includes 4 open source models. The American company boasted that its Gemma 4 26B and Gemma 4 31B Dense models were more powerful than the Chinese company's Deepseek-v3.2-exp-thinking model. We will now have to see how DeepSeek V4 Pro positions itself compared to these two Google open source models.

In any case, China continues to position itself as a leader in open source AI, while also becoming the world champion in the recycling of electronic devices.