DeepSeek launches DeepSeek-V4 preview AI models

Beijing-based artificial intelligence startup DeepSeek released preview versions of two new chatbots, DeepSeek-V4-Pro and DeepSeek-V4-Flash, on Friday, April 24, 2026, in Beijing, the company announced. The launch places DeepSeek more directly in competition with major U.S. AI developers, including OpenAI and Google, as global demand for advanced chatbot tools continues to expand.

 

According to the company announcement, DeepSeek-V4-Pro is positioned as a high-end model and is reported to outperform all rival open models in mathematics and coding. DeepSeek also said the model ranks behind only Google’s Gemini 3.1-Pro in world knowledge, indicating that the company is benchmarking its latest release against leading proprietary systems as well as open alternatives.

 

DeepSeek said its second preview model, DeepSeek-V4-Flash, is designed to deliver reasoning performance comparable to the Pro version while prioritizing speed and cost. The company described Flash as offering faster response times and more cost-effective pricing, a combination that can be relevant for developers and businesses that need lower latency and tighter operating budgets.

 

Both DeepSeek-V4-Pro and DeepSeek-V4-Flash are open-source, the company said, meaning developers can use and modify the models freely. That approach can broaden adoption by lowering barriers for software teams and researchers, while also intensifying competition in the open-model ecosystem where performance claims are often tested quickly by independent users.

 

The new previews follow DeepSeek’s earlier release of DeepSeek-R1 in January 2025. DeepSeek said DeepSeek-R1 showed capabilities comparable to ChatGPT and Gemini, and the company also claimed it built that model with less than $6 million in computing costs, a figure it described as far below typical Silicon Valley AI development budgets.

 

DeepSeek’s prior model also drew regulatory and policy scrutiny in multiple jurisdictions. DeepSeek-R1 faced restrictions in several countries and regions, including multiple U.S. states, Australia, Taiwan, South Korea, Denmark, and Italy, with concerns cited around data protection and national security.

 

For global markets, the release highlights how quickly open-source AI offerings are evolving and how pricing and performance claims can influence developer choices across regions. At the same time, the earlier restrictions underscore an ongoing uncertainty for cross-border deployment: even when models are technically accessible, policy limits tied to security and privacy can shape where and how they are used.