Chinese AI startup DeepSeek, originated from High-Flyer Capital Management, has launched DeepSeek-V4, a groundbreaking 1.6-trillion-parameter Mixture-of-Experts (MoE) AI model. Offered under an open-source MIT License, it delivers near state-of-the-art intelligence comparable to top-tier closed-source models like OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.7 but at roughly one-sixth the cost. DeepSeek-V4 offers vast improvements in efficiency, long-context handling with a 1 million token window, and stability through novel architectural innovations like Manifold-Constrained Hyper-Connections (mHC) and a Hybrid Attention Architecture. The model is priced competitively via API, potentially transforming the economics of AI deployment for enterprises by making advanced AI capabilities more accessible and affordable. It supports multiple reasoning modes and is compatible with NVIDIA GPUs and Huawei’s Ascend NPUs, promoting sovereignty in AI technology infrastructure. Community and industry responses highlight the model’s disruptive potential, placing pressure on established AI providers to justify their pricier offerings while broader access promises significant benefits for AI integration worldwide.
Back
