Unverified50% confidenceOpinionExact time
M3.1 Flash's activation ratio is about 5% (23B/428B), which is a relatively aggressive design among current open-source and semi-open-source models.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedMaximum output tokens: MiniMax M3.1 Flash is 512K, DeepSeek V4.1 Flash 1017 is 384K, and Claude Haiku 5.5 is 128K.73% similarUnverifiedQwen 3.8 Flash has a parameter count of 180B, which is half that of GLM 5.3 Flash (320B).72% similarUnverifiedThe open-source project StraightA claims to be able to run the 125B-parameter Qwen 3.8 Flash Next model on a consumer-grade PC with more than 12G of VRAM, provided the system is equipped with 35 to 55G of RAM.71% similarUnverifiedMiniMax M3.1 Flash uses a MoE architecture with approximately 428B total parameters and approximately 23B activated per token.70% similarUnverifiedGLM 5.3 Flash has a total model size of 320B, with activated parameters reaching 18B.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1068672API
curl https://kongchang.com/api/v1/knowledge/claims/1068672MCP
get_claim(id=1068672)