New Delhi, Apr 24 (New Delhi): DeepSeek has introduced preview versions of its new V4 series AI models, signalling intensifying competition in the global artificial intelligence race.
Model Launch And Technical Claims
The China-based startup released the V4 Flash and V4 Pro preview models, stating that both versions focus on improved reasoning capabilities and support for agent-based tasks. According to details shared on developer platforms, the models can handle context lengths of up to one million tokens, enabling them to process longer prompts and conversations.
DeepSeek also outlined a new design approach called Hybrid Attention Architecture, which aims to improve how AI systems retain and reference information across extended interactions. The company indicated that the upgrades include architectural and optimisation changes intended to enhance coding and reasoning performance.
Background And Market Impact
The announcement follows the company’s earlier release of the R1 model, which drew global attention for delivering competitive results at reportedly lower development costs. That release contributed to renewed discussions among investors and technology firms about the economics of large-scale AI development. Spending in the sector continues to rise. Technology companies in the United States are expected to invest close to $650 billion in data centres and AI infrastructure by 2026, reflecting sustained demand for advanced computing resources.
Rising Competition And Regulatory Scrutiny
Competition in China’s AI ecosystem has accelerated in recent months, with companies such as Alibaba Group Holding Ltd. and Baidu Inc. expanding their AI offerings. Firms including ByteDance Ltd., Zhipu AI, and MiniMax have also rolled out updates to their models ahead of April, reflecting heightened competition in the domestic market. At the same time, DeepSeek’s growth has drawn scrutiny from U.S. officials and technology companies. Concerns have been raised over the possible use of model distillation techniques, where one system trains using outputs from another to replicate capabilities. Both OpenAI and Anthropic have previously flagged suspected attempts linked to such practices. Questions have also emerged around access to advanced chips produced by Nvidia Corporation, which are subject to export controls. Authorities have examined whether restricted hardware could have been used in developing certain AI models.
Also Read: OpenAI Appoints Nitin Bawankule to Lead Enterprise Sales in India as AI Adoption Scales
