Unverified50% confidenceBenchmarkExact time
Artificial Analysis独立评测中V4.1 Flash智能指数得39分，超过更大的V4 Pro的36分，测得约232 tokens/秒
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
9/27/2026
First Seen
Valid until: 10/11/2026
Sources
Related Entities
Related Claims
UnverifiedV4.1-Flash在某编程测试维度的得分达到74.2分，相比V4-Flash提升了将近20分77% similarUnverifiedArtificial Analysis给Muse Glimmer 30B打出35分的智能指数，相比Llama 4 Maverick是巨大飞跃71% similarUnverifiedV4.1 Flash提供低、中、最高三档推理强度，八项推理基准平均分从67.1涨到76.3，代价是输出Token增加约2.5倍70% similarUnverified在一道三样本数学题上，DeepSeek V4.1 Flash达到74分，与Opus 5.0持平，超过GLM的67分和V4 Pro的62分70% similarUnverifiedAgnes-3.0-Flash 的全局注意力层采用 24 个 query head 与 4 个 KV head 的 6:1 GQA，head dim 为 256，对 q 和 k 施加 RMS-norm，输出经过 sigmoid 门控66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/954589API
curl https://kongchang.com/api/v1/knowledge/claims/954589MCP
get_claim(id=954589)