Unverified50% confidenceFactExact time
模型最终参数量为13,693,824个，落在10-30百万目标区间内
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
Unverified分析推测该模型是2到4万亿参数规模的MoE模型，每个Token仅激活约1500亿参数74% similarUnverified推理模型在输出最终答案前内部思维链可能消耗3000至20000个Token，远超普通模型的约500个Token69% similarUnverified1000亿参数的模型处理单次推理请求需约80GB显存，百亿参数级别模型仅需约20GB68% similarUnverified一个千亿参数级别旗舰模型每次推断所需GPU资源可能比百亿参数模型高出10至20倍68% similarUnverifiedGrok 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66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/827402API
curl https://kongchang.com/api/v1/knowledge/claims/827402MCP
get_claim(id=827402)