DeepSeek Unveiled V4: Chinese Open-Source Model Challenges OpenAI and Google

DeepSeek is trying to shift the AI market again not with loud promises, but with a release that hits several pain points in the industry at once. On April 24, 2026, the Chinese company released a preview of the V4 series and announced that the new open-source model is capable of competing with the closed systems of OpenAI, Google, and Anthropic. For the market, this is important news not only because of the models themselves, but also because DeepSeek is betting on coding, agentic scenarios, and its own Chinese computational base.

The series includes two versions: DeepSeek-V4-Pro and DeepSeek-V4-Flash. The first is a flagship model with 1.6 trillion parameters and 49 billion active parameters, the second is a more compact variant with 284 billion parameters and 13 billion active ones.

Both support context up to 1 million tokens, and the weights are published under the MIT license, meaning the company continues its course toward maximally open distribution of its models. According to DeepSeek's own description, the V4 architecture has been significantly redesigned: the company has focused on efficiency of long context, training robustness, and stronger post-training alignment for different types of tasks. The main focus in V4 is programming.

DeepSeek directly states that the new generation has become noticeably stronger at coding, and this ability now forms the foundation of AI agents that write, fix, and run code. Against the backdrop of growing tools like ChatGPT Codex and Claude Code, this is no longer a secondary benchmark but one of the key indicators of how suitable a model is for real-world work. In the tests published by DeepSeek, V4-Pro Max mode shows 93.

5% on LiveCodeBench and 80.6% on SWE Verified, and on some tasks it approaches the best closed models. The company itself carefully formulates the result like this: V4 doesn't necessarily lead everywhere, but noticeably closes the gap between open and closed systems on complex reasoning tasks and in agentic scenarios.

A separate part of the story is hardware. DeepSeek emphasizes V4's compatibility with the Huawei Ascend line, and this makes the release politically and industrially more significant than just a routine model update. A year ago, almost all talk about advanced AI models revolved around American chips and especially Nvidia.

Now one of the most prominent Chinese players is trying to show that a competitive model can not only be trained in China, but also be integrated into a local stack — from training and inference to optimization for domestic accelerators. For Huawei, this is an important signal: its AI infrastructure is becoming not a backup option, but a platform for first-tier releases. The context for this announcement is also strong.

A year ago, DeepSeek already shook the American AI market with the R1 model, claiming it was able to train it for significantly less money than industry leaders in the US spend. With V4, the company has not yet disclosed the training cost and does not clarify the full hardware stack used, so many questions remain around the release. Against this backdrop, political pressure continues: American officials previously accused DeepSeek of using prohibited Nvidia chips, and Anthropic claimed the company abused access to Claude to improve its own products.

Proven transparency is still lacking here, and this will affect how V4 is perceived outside of China. The main conclusion is simple: V4 is not just another model in the long stream of AI releases. DeepSeek is trying to prove three things at once: the open-source approach is still capable of catching up with frontier-class models, coding is becoming a central discipline for LLMs, and the Chinese AI ecosystem increasingly wants to rely less on American hardware and closed APIs.

If the announced results are confirmed in practice, the market will receive not only a new strong open-source player, but also another argument in favor of the fact that the AI race is now happening simultaneously at the level of models, tools, and chips.