Unverified50% confidenceFactExact time
1.21亿参数的完整Neural 3模型实际计算量约等于一个5000万参数模型
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
Unverified对于1.6万亿总参数、490亿激活参数配置，实际推理时只有约3%的参数参与计算77% similarUnverified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76% 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20GB76% similarUnverified通义千问3.8约有两万亿参数，Kimi约有两万亿参数74% similarVerified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Token73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977851API
curl https://kongchang.com/api/v1/knowledge/claims/977851MCP
get_claim(id=977851)