Home Page
Deep Impact on Three Levels
The 'computing power as a barrier' is broken
💡 Key Facts of DeepSeek
Achieving stronger capabilities with fewer active parameters
💡 Key facts about DeepSeek
· V3 training cost: approximately5.57 million dollars(2,048×H800, 54.2 days)
5.57 million US dollars
589 billion US dollars
Consultation and Cooperation →
· The market value of NVIDIA evaporated approximately589 billion dollars
Deep-seated impacts on three levels
Impact one: The reshaping of the API pricing system
Impact Two: Diversification of Technical Routes
- Baidu Wenxin Yiyan: Multiple model call prices dropped to free
- Model Capabilities Tend to Homogenize
- Technical Insights
- Reinforcement Learning Drives Inference Ability
Insights for Industry Implementation:
Impact two: Diversification of technical routes
The 'Computing Power as Barrier' is Broken
- MoE is superior to dense modelsAchieve stronger capabilities with fewer active parameters
- Reinforcement learning drives inference capabilitiesPrivate Deployment Becomes Possible
- Engineering innovation makes up for hardware gapsEngineering Innovation Makes Up for Hardware Gap
Impact three: Acceleration of the open-source ecosystem
Nine
Insights for the industry
Impact One: The Re-shaping of the API Price System
Insights for industry implementation:
- Model capabilities tend to homogenizeDeepSeek Effect
- Data value stands outMoE Outperforms Dense Models
- Private deployment becomes possibleData Value Prominently Highlighted
- China's AI independent innovation gains global recognitionInsights for the Industry
China's AI independent innovation gains global recognition