DeepSeek V4 preview released, Chinese startup unveils V4 model

Chinese AI startup DeepSeek has released a preview of its long-awaited V4 large language model, introducing two initial variants — V4-Pro and V4-Flash — via an announcement on its official WeChat channel. The preview release began on April 24, 2026 and marks the first public availability of the next-generation family.

Initial technical disclosures and reporting indicate the V4-Pro variant can scale to roughly 1.6 trillion parameters while V4-Flash is positioned as a lighter, cost-efficient alternative. Reports also point to significantly expanded context windows and strengthened coding and reasoning capabilities; some sources reference context capacity claims reaching up to one million tokens, though full technical papers and model cards were to be published incrementally.

The preview is likely to have immediate implications for Chinese cloud providers and domestic chipmakers. Earlier reporting suggested DeepSeek has prioritized compatibility and early optimization with China-native accelerators, including Huawei designs, which could give local hardware vendors an edge in accelerating model deployment and reducing reliance on Western accelerators. Competitive pricing of V4 could also pressure international cloud and model providers.

In broader context, DeepSeek’s prior releases drew global attention for offering high performance at low cost, and V4’s release underscores how rapidly the Chinese AI ecosystem is advancing. The company has faced scrutiny over data and training practices in past months, and regulatory and reputational considerations may shape how widely and quickly V4 is adopted outside China.

Market analysts say the V4 preview could trigger short-term re-rating across technology suppliers, cloud platforms and AI infrastructure vendors as investors reassess competitive dynamics. The market will closely watch DeepSeek’s forthcoming technical reports, pricing details and partner integrations to verify performance claims; these factors will determine whether V4 materially shifts cost and capability benchmarks in the LLM market.