DeepSeek Unveils V4-Pro and V4-Flash: A Year After Its 'Sputnik Moment'
April 24, 2026 - 7:06 am
The Hangzhou startup released preview versions of both models on Hugging Face on Friday. V4-Pro claims top performance in coding and math among open models, trailing only Gemini 3.1-Pro for world knowledge, and falls “marginally short” of GPT-5.4 and Gemini 3.1-Pro, a gap DeepSeek estimates at 3 to 6 months.
Both models are open-source.
DeepSeek, the Hangzhou-based AI startup that shocked Silicon Valley with its R1 model in January 2025, released preview versions of its latest flagship models on Friday, exactly one year later. The company posted DeepSeek-V4-Pro and DeepSeek-V4-Flash on Hugging Face, positioning them as the most powerful open-source AI platform available and a direct challenge to rivals like OpenAI and Anthropic.
Both models inherit the open-source model architecture from their predecessors, allowing developers to use and modify the source code freely. The key technical advancement in V4 is the Hybrid Attention Architecture, which DeepSeek claims enhances context retention across long conversations. Combined with a 1-million-token context window, sufficient to process entire codebases or book-length documents in a single prompt, the architecture supports agentic and long-horizon reasoning tasks where previous models struggled.
The Flash variant prioritizes speed and cost efficiency, while the Pro variant focuses on peak performance.
DeepSeek’s own benchmarks place V4-Pro as the top open-source model in coding and math, second only to Gemini 3.1-Pro in world knowledge. Compared to the current closed-source models, OpenAI’s GPT-5.4 and Gemini 3.1-Pro, DeepSeek acknowledges that V4-Pro lags slightly, by approximately 3 to 6 months, a candid assessment reflecting either unusual intellectual honesty or a strategic move to set conservative expectations for independent evaluation.
The chip story adds a geopolitically charged layer. DeepSeek collaborated with Chinese AI chipmakers Huawei and Cambricon to optimize V4 for their latest hardware, according to Reuters, citing The Information. The company chose not to grant Nvidia or AMD early access for optimization, reversing standard industry practice. Running a frontier-class model at this scale on Huawei’s Ascend chips would be a significant proof of concept for China’s domestic AI hardware supply chain, currently under US export restrictions.