Unverified50% confidenceFactExact time
HY3's Top-8 routing activation ratio is approximately 7% (21 billion / 295 billion).
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
腾讯HY3开源模型深度实测：轻量MoE硬刚旗舰级性能
bilibiliAI_Express7/8/2026
Related Claims
UnverifiedHy3 adopts a Mixture of Experts (MoE) architecture with a total parameter count of 295 billion (295B), activating only 21 billion (21B) parameters per inference.63% similarUnverifiedHY3 internally contains 192 experts and uses a Top-8 routing mechanism.62% similarUnverifiedIn the proprietary benchmark PinBash 3 test, SOL scored approximately 78.6%, TERRA approximately 62.9%, and LUNA approximately 44.3%.56% similarUnverifiedKimi K3's 2.8 trillion parameter scale surpasses the previously largest open-weight model, DeepSeek-V3, which had 671 billion parameters.55% similarUnverifiedHy3's score on the MRCR (Multi-turn Referential Coreference Resolution) benchmark jumped from 42.9% to 75.1%.54% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/606579API
curl https://kongchang.com/api/v1/knowledge/claims/606579MCP
get_claim(id=606579)