Unverified50% confidenceFactExact time
Muse-Glimmer-30B参数规模为30B级别（实测架构约28B），支持128K上下文，并在发布时提供本地量化版本
1
Sources
50%
Confidence
Long-term
Relevance
8/16/2026
First Seen
Sources
Related Claims
Partially VerifiedMuse Glimmer全精度运行需要55GB显存，通过4-bit量化后压缩到20GB以内，可在24GB显存的消费级显卡上运行71% similarUnverifiedQwen3-235B-A22B采用MoE架构，235B总参数中每次只激活22B70% similarUnverifiedQwen3.6 27B支持262K上下文长度，34B 80层版本保持相同的上下文长度不变69% similarVerified32B参数规模的模型在量化后通常可以压缩到约16-20GB显存占用，一张NVIDIA RTX 4090（24GB显存）就能运行69% similarUnverifiedMuse Glimmer的原生上下文长度仅为128K67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/757561API
curl https://kongchang.com/api/v1/knowledge/claims/757561MCP
get_claim(id=757561)