Expired50% confidenceFactExact time
Kimi 2.5 computes only approximately 30 billion parameters per forward pass, despite having a total parameter count of 1 trillion.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Cursor Composer2 Training Revealed: A Complete Guide to Distributed Reinforcement Learning Engineering
bilibili豚工智能6/6/2026
Related Claims
UnverifiedAnalysis speculates that the model is an MoE model with 2 to 4 trillion parameters, activating only approximately 150 billion parameters per token.71% similarUnverifiedA 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.69% similarUnverified苹果的端侧模型号称有200亿参数，采用MoE式部分激活机制，每次推理只激活10亿到40亿参数66% similarUnverifiedIf a MoE model activates only approximately 1/8 of its parameters per inference, the actual computational overhead may be equivalent to that of a dense model with tens of billions of parameters.65% similarUnverified目前能在手机端流畅运行的模型参数量通常在30亿至70亿级别，与云端数千亿参数的旗舰模型存在数量级差距65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/51431API
curl https://kongchang.com/api/v1/knowledge/claims/51431MCP
get_claim(id=51431)