DeepSeek launches V4 AI models to challenge global leaders in advanced AI

Chinese AI firm DeepSeek has unveiled its V4 model series, including Pro and Flash variants, aiming to rival leading US systems with high performance, cost efficiency, and support for ultra-long context processing capabilities.

DeepSeek has introduced its latest artificial intelligence models under the V4 series, marking a significant step in its efforts to compete with leading global players in the rapidly evolving AI landscape. The Hangzhou-based company said the new models are designed to deliver performance comparable to top-tier proprietary systems developed by firms such as OpenAI and Google DeepMind.

The release includes two variants—V4-Pro and V4-Flash—representing the company’s most advanced models to date. The larger V4-Pro model features 1.6 trillion parameters, while the more lightweight V4-Flash version is built with 284 billion parameters. Both models are open-source and aim to strike a balance between computational power and cost efficiency, a key factor in today’s competitive AI market.

Focus on scale, efficiency and long-context processing

A key highlight of the V4 series is its ability to process extremely large volumes of information. Both models support a context window of up to one million tokens, enabling them to handle complex, long-form tasks such as advanced coding, research analysis, and multi-step problem-solving.

The company attributed this capability to architectural improvements that significantly enhance computational efficiency when dealing with long sequences of data. According to DeepSeek, this advancement could open new possibilities in areas requiring sustained reasoning over extended inputs, while also supporting emerging AI paradigms such as continuous learning systems.

In benchmark evaluations, the company stated that V4-Pro outperforms most open-source alternatives and comes close to matching leading closed-source systems, including models like Google’s Gemini series.

Hardware constraints and industry challenges

While the models demonstrate strong performance capabilities, DeepSeek acknowledged limitations related to computing infrastructure. The company did not disclose the exact hardware used for training but indicated that its systems are compatible with both Nvidia and Huawei chip architectures.

The firm noted that broader access to advanced hardware will be crucial for scaling performance and reducing operational costs. It expects improvements later in the year as newer chipsets, including Huawei’s upcoming systems, become more widely available.

The launch comes at a time when Chinese AI companies are navigating restrictions on access to high-end semiconductors, particularly advanced graphics processing units. Despite these challenges, DeepSeek’s latest release underscores the growing ambition of domestic players to build competitive, large-scale AI models capable of rivaling global benchmarks.

With the V4 series, DeepSeek is positioning itself as a key contender in the international AI race, focusing on scalability, efficiency, and open innovation.