Verified65% confidenceFactExact time
GPT-4采用混合专家（MoE）架构，估计总参数量达1.8万亿，每次推理实际激活参数约为总量的1/8
3
Sources
65%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedEach inference requires loading model weights into GPU memory and generating responses token by token; GPT-4-level model parameter counts are typically on the order of hundreds of billions.85% similarUnverifiedGPT-3拥有1750亿参数，用进化策略进行有效采样所需的函数评估次数随参数维度呈指数级增长（维度灾难）78% similarUnverifiedGPT-4的参数量据估计超过1万亿，训练过程涉及的浮点运算次数达到10²⁴数量级78% similarUnverifiedModel parameter scale has grown from 340 million in BERT to an estimated over one trillion in GPT-4.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/564525API
curl https://kongchang.com/api/v1/knowledge/claims/564525MCP
get_claim(id=564525)