DeepSeek Launches New DeepSeek-V4 AI Model with 1 Million Word Context Window
A Chinese startup matches top-tier US AI with a fraction of the cost...
Chinese startup DeepSeek released DeepSeek-V4 on Friday, over a year after its R1 model shocked the AI world by matching US rivals at a fraction of the cost. The new model features an ultra-long context window of one million words, enabling it to process entire books or lengthy documents in a single pass. DeepSeek-V4 comes in two versions: the full V4-Pro with 1.6 trillion parameters, and the leaner V4-Flash with 284 billion parameters, designed for efficiency and lower cost.
In benchmark tests, V4-Pro leads all other open-source models in world knowledge, agent capabilities, and reasoning, and is only slightly outperformed by Google's top-tier closed-source Gemini-Pro-3.1. The model is optimized for popular AI agent tools like Claude Code, OpenClaw, and CodeBuddy. DeepSeek's open-source approach contrasts with proprietary models from OpenAI and Google, and has driven widespread adoption by Chinese municipalities, healthcare institutions, and financial firms. The announcement came as Meta and Microsoft reportedly plan workforce cuts to invest more heavily in AI.
- DeepSeek-V4 features a 1 million word context window, allowing processing of entire documents in one go.
- V4-Pro has 1.6 trillion parameters; V4-Flash has 284 billion parameters for cost efficiency.
- The model leads open-source benchmarks and is optimized for AI agent tools like Claude Code and CodeBuddy.
Why It Matters
DeepSeek's open-source, cost-effective model intensifies the US-China AI race, challenging proprietary giants and reshaping global competition.