China's DeepSeek previews new AI model to rival US giants
What's the story
Chinese artificial intelligence (AI) start-up DeepSeek has launched its latest AI model, the V4. The new open-source model is said to be on par with closed-source models from US tech giants such as OpenAI and Google DeepMind. The release comes in two versions: the 1.6 trillion parameter DeepSeek-V4-Pro and the 284 billion parameter DeepSeek-V4-Flash.
Technological advancement
Model supports context length of 1 million tokens
The company claims that the DeepSeek-V4 series marks a major leap in computational efficiency for processing ultra-long sequences. This innovation allows support for a context length of one million tokens, paving the way for deeper research into long-horizon tasks and future paradigms like online learning. The exact hardware used to train these models remains undisclosed but is said to include software components compatible with NVIDIA and Huawei chips.
Market impact
DeepSeek-V4 tuned for popular agent tools
DeepSeek's latest release comes more than a year after the launch of its R1 reasoning model, which made waves in global tech markets with its surprising performance and efficiency. The company claims that V4 performs well against domestic competitors, especially in agent-based tasks, knowledge processing, and inference. It has also been optimized for popular agent tools such as Anthropic's Claude Code and OpenClaw.
Company evolution
DeepSeek's journey so far
Founded in 2023, DeepSeek gained attention in late 2024 with its free, open-source V3 model. The company trained this model with less powerful chips and at a fraction of the cost of models built by OpenAI and Google. In January 2025, it launched the R1 reasoning model that matched or outperformed many leading LLMs worldwide.