Shares in several Chinese artificial intelligence firms moved lower on Friday after DeepSeek published preview builds of its new V4 model, a development market participants said raised the sector's competitive stakes.
DeepSeek issued preview versions of V4 offering two distinct configurations - a high-performance "Pro" edition and a more affordable "Flash" variant designed to encourage broader use among developers. The company highlighted that V4 supports an ultra-long context window of up to 1 million tokens and includes enhanced agent-style functionality, positioning it to handle more complex enterprise and research workflows.
The firm released the preview as an open-source project. That decision aligns with a low-cost, high-access approach that the company says will influence pricing dynamics across China’s expanding AI market.
Market moves were visible in Hong Kong trading: Zhipu AI, listed as Knowledge Atlas Tech (HK:2513), fell 8% to HK$946 by 04:57 GMT. Shares of MiniMax Group (HK:0100) declined 4.7% to HK$818.5.
The preview release is intended to collect user feedback prior to a wider rollout. Observers have pointed to the model’s strong benchmark performance and reported efficiency gains as reasons the preview has attracted attention.
Context and market response
DeepSeek’s open-source emphasis and the availability of both premium and lower-cost model builds were presented as strategic moves to broaden adoption while altering cost structures for AI services. The preview’s technical features - notably the extended context window and agent-style enhancements - were highlighted in the company’s materials as relevant for advanced enterprise and research applications.
In Hong Kong equity trading, the announcement coincided with measurable declines in the quoted prices of two AI-related firms, reflecting investor sensitivity to shifts in competitive positioning and pricing expectations within the sector.
Further developments
The V4 preview is a sounding board for user feedback ahead of a fuller release. Attention around benchmark outcomes and efficiency indicates the market is monitoring both technical and commercial implications of the model as they unfold.