DeepSeek has released preview versions of its new flagship Artificial Intelligence models, the V4 Flash and V4 Pro, marking its most significant launch since the debut of R1. The company described the new series as its most powerful open-source platform and framed the rollout as a direct challenge to rivals including OpenAI and Anthropic. DeepSeek said the new models deliver top-tier performance in coding benchmarks and major gains in reasoning and agentic tasks.
The V4 Flash and V4 Pro come with several architecture upgrades and optimisation improvements. They can operate with a million-token context length, according to the company’s post on Hugging Face. DeepSeek highlighted a technique it calls Hybrid Attention Architecture, saying it improves an Artificial Intelligence platform’s ability to remember queries across long conversations. The release underscores DeepSeek’s effort to build on the momentum created by R1, which helped establish the Hangzhou-based startup as a serious force in open-source model development.
The V4 arrives more than a year after the Hangzhou-based startup ignited a trillion-US dollar stock market sell-off with the release of the R1, an open source model that mimics the process of human reasoning. The R1 rivalled the performance of cutting-edge Artificial Intelligence systems from companies like OpenAI but was purportedly built for a fraction of the cost. Almost overnight, some tech firms and investors began rethinking the wisdom of pouring billions of US dollars into Artificial Intelligence development. Those outlays have since sprung back, as American technology giants are projected to invest around US? billion in 2026 on Artificial Intelligence infrastructure and data centres.
DeepSeek’s rise has also intensified competition in China. Tech leaders from Alibaba Group Holding to Baidu have pushed low-cost Artificial Intelligence services into the market, while rivals including ByteDance, Zhipu and Minimax moved to refresh their own models in the weeks leading up to April. At the same time, DeepSeek faces growing scrutiny from American technology leaders and government officials. The concerns centre on allegations that the company may have used distillation, in which one model relies on another model’s output for training, and on whether it may have gained access to banned Nvidia Artificial Intelligence chips. OpenAI and Anthropic have both alleged they detected such attacks from DeepSeek, while US officials began probing the chip issue last year.
