BREAKING
DeepSeek R1 Sparks $589B Nvidia Rout
0B
Value wiped out
0%
Share drop
How DeepSeek Built R1
1DeepSeek-V3 base
2RL for reasoning
3Open MIT weights
0M
V3 training cost
0$
Per M input tokens
0$
Per M output tokens
R1 vs OpenAI o1
DeepSeek R1Open weights
Strong on AIME, MATH-500
MoE efficiency
Distilled 1.5B to 70B
OpenAI o1Closed API
Leading reasoning family
Higher cost
Some tasks still ahead
Efficiency Reframes the AI Debate
AI NEWS BLITZ
A low-cost Chinese AI model triggered one of the biggest stock shocks ever recorded.