DeepSeek launches V4-Pro and V4-Flash, one year after historic breakthrough

The Hangzhou-based AI startup that sent shockwaves through Silicon Valley in early 2025 with its R1 model has returned with its next generation of frontier technology. DeepSeek released preview versions of DeepSeek-V4-Pro and DeepSeek-V4-Flash on Hugging Face this past Friday, marking roughly one year since its historic breakthrough.

The company is positioning these models as the most powerful open-source AI platform currently available, directly challenging offerings from OpenAI and Anthropic. True to form, both releases remain open-source, granting developers full access to modify and use the source code.

The standout technical innovation in V4 is the Hybrid Attention Architecture, which DeepSeek claims significantly enhances the model’s ability to maintain context over extended conversations. Combined with a 1-million-token context window,enough to digest an entire codebase or a book-length document in a single prompt,this architecture is built for agentic and long-horizon reasoning tasks. In previous models, quality tended to degrade as context length grew, but DeepSeek says V4 overcomes that limitation. The Flash variant prioritizes speed and cost efficiency, while the Pro version is optimized for peak capability.

According to DeepSeek’s internal benchmarks, V4-Pro leads all open-source models in coding and mathematics. In world knowledge, it trails only Google’s closed-source Gemini 3.1-Pro. When compared to the current closed-source frontier,OpenAI’s GPT-5.4 and Gemini 3.1-Pro,the company admits V4-Pro falls only “marginally short.” In a candid self-assessment, DeepSeek estimates its “developmental trajectory” lags behind state-of-the-art frontier models by approximately 3 to 6 months.

That level of transparency is rare in AI model launches, which typically emphasize areas where the new model outperforms competitors. By publishing a gap estimate rather than claiming parity, DeepSeek is either demonstrating unusual intellectual honesty or strategically managing expectations ahead of independent evaluations.

The geopolitical dimension of this release is impossible to ignore. According to Reuters, citing The Information, DeepSeek worked with Chinese AI chipmakers Huawei and Cambricon to optimize V4 for their latest hardware. Notably, the company did not give Nvidia or AMD early access for optimization, a reversal of the standard industry practice where Western chipmakers are typically first to receive new model weights for performance tuning.

Running a frontier-class model at this scale on Huawei’s Ascend chips,rather than on Nvidia H100s or H200s,would represent a significant proof of concept for China’s domestic AI hardware supply chain, which has been under US export restrictions since October 2022. The V4 release does not erase that geopolitical constraint, but it tests its limits in a commercially visible way.

The timing is deliberate. DeepSeek-R1’s launch in January 2025 erased roughly $600 billion from Nvidia’s market capitalization in a single day, as investors recalibrated their assumptions about the compute required to build frontier AI. Marc Andreessen called it “AI’s Sputnik moment.” The practical claim was that a Chinese lab had matched OpenAI’s best reasoning model while spending less than $6 million on compute,a figure some analysts disputed but which nonetheless reset global assumptions about the economics of frontier AI development.

V4 arrives into a very different market. OpenAI released GPT-5.5 on the same day. Anthropic is now valued at $1 trillion on secondary markets. The US-China AI competition has become an explicit dimension of trade and technology policy. DeepSeek’s second act is landing in a much more contested arena than its first.

Both models are preview releases, not final production versions. Independent benchmarking has not yet been completed, and DeepSeek’s own benchmarks should be treated as preliminary until third-party evaluation confirms them. The same caveat applied to R1, whose claims were broadly validated by external testing within days of release. Whether V4 holds up to the same scrutiny will be clear within the week.