A year after rattling Silicon Valley with its technology, China’s DeepSeek rolled out preview versions of a new flagship AI model. The Chinese startup unveiled the V4 Flash and V4 Pro series, claiming top-tier performance in coding benchmarks and major advancements in reasoning and agentic tasks.
DeepSeek highlighted a technique it described as Hybrid Attention Architecture, which it said improves the platform’s ability to retain context across extended conversations. It also introduced a 1 million-token context window — enabling entire codebases or long documents to be submitted as a single prompt.
The V4 release arrives as US AI labs including OpenAI and Anthropic continue their own aggressive release cycles — and at a moment when enterprise AI buyers are increasingly evaluating open-source models as viable alternatives to proprietary platforms for cost and sovereignty reasons.
