Unverified50% confidenceFactExact time
Qwen3.8 Flash Next has a sparsity rate of approximately 95%, using less than 5% of its parameters per inference.
1
Sources
50%
Confidence
Long-term
Relevance
9/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIt is widely speculated in the industry that Flash's parameter scale is approximately one-third to one-half that of Pro.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70% similarUnverified单条推理通常只能占用GPU SM利用率的5%～15%，而合理的批次大小可将利用率提升至60%～90%69% similarVerifiedThe Mixture of Experts (MoE) architecture activates only approximately 10%–30% of total parameters during inference, significantly reducing FLOPs and GPU memory usage per inference67% similarUnverifiedThe per-token inference cost of Flash models is typically one order of magnitude lower than that of Pro-level models.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/843529API
curl https://kongchang.com/api/v1/knowledge/claims/843529MCP
get_claim(id=843529)