The Attention Hybrid MoE Architecture is the Future. Now, AI Labs Should Dedicate Resources to Improve Long Context Recall Capabilities.