DeepSeek V4: Why an Open AI Model Is So Cheap to Run

DeepSeek V4 is an open-weight frontier AI model that is unusually cheap to run, and this video explains exactly why using the idea of Mixture-of-Experts.

You will learn what an open-weight model really is (and why it is not the same as open source), how a 1.6-trillion-parameter model can fire only about 3 percent of itself on each word, and how a small router picks a handful of experts per token. We unpack why that decoupling makes the model cheap to run while still smart, why it still needs serious hardware to load into memory, and the crucial difference between training cost and inference cost. We also read its benchmark scores like a scientist, separating self-reported numbers from independent tests, and clear up the biggest myths people repeat about DeepSeek.

Chapters:
0:00 The trillion-parameter puzzle
0:55 What DeepSeek V4 is
2:17 Why open weights matter
3:24 The market shock of 2025
4:48 Mixture-of-Experts explained
7:10 Smart experts and long context
8:36 Why it is cheap to run
11:11 Cheap to build or to run
14:52 Reading benchmarks honestly
17:34 Myths and what it means

📺 More AI, explained simply: https://www.youtube.com/playlist?list=PLM5VvmudKYKcyt09H05mKtarxRNXGTtoN

Subscribe to @HowAIWorksHQ for clear, honest explanations of how AI actually works.

DeepSeek V4, DeepSeek explained, open weights, open source AI, Mixture of Experts, MoE, AI economics, inference cost, frontier model, LLM explained, what is DeepSeek

#DeepSeek #DeepSeekV4 #OpenWeights #MixtureOfExperts #AIExplained #LLM #OpenSourceAI #HowAIWorks

Видео DeepSeek V4: Why an Open AI Model Is So Cheap to Run канала How AI Works!