Unverified50% confidenceFactExact time
Grok 1 的3140亿总参数中每次推理仅激活约25%，实际推理算力接近650亿参数规模的密集模型
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedA model with 100 billion parameters requires approximately 80 GB of VRAM to process a single inference request, while a model with tens of billions of parameters requires only about 20 GB.74% similarUnverifiedAnalysis speculates that the model is an MoE model with 2 to 4 trillion parameters, activating only approximately 150 billion parameters per token.70% similarUnverifiedBefore outputting a final answer, a reasoning model's internal chain of thought may consume 3,000 to 20,000 tokens, far exceeding the approximately 500 tokens of an ordinary model.70% similarUnverified参数量在1亿以下的轻量模型经领域数据微调后可达95%以上意图分类准确率，单次推理延迟控制在5-20ms以内69% similarUnverifiedModels with larger parameter counts and more reasoning steps may consume 10–50 times the GPU compute of lightweight models.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566579API
curl https://kongchang.com/api/v1/knowledge/claims/566579MCP
get_claim(id=566579)