Tencent Hunyuan released Hy4 Flash, a 224B-parameter (12B activated) multimodal model with 1M context, priced at RMB 0.5 per million input tokens. Through LatentMoE, Engram, 4-bit quantization and megakernel optimizations, its KV cache is one sixth of Hy4's and per-task cost drops to one twentieth, while slightly outperforming the 770B Hy4 preview. Hunyuan plans to open-source it with frequent iterations.
Termsslug: latentmoe
LatentMoE
潜在混合专家
MoE variant performing expert routing in latent space to improve sparse activation efficiency.
Curated Intelligence
InfoQ 中文T2·Models
78 pts
IT之家 · 智能时代T2·Models
78 pts
Tencent Hunyuan officially launched Hy4 Flash, a 224B-parameter model activating 12B per token with multimodal input and 1M-token context. Through coordinated pretraining, post-training and inference optimizations including hybrid attention, LatentMoE, Engram lookup parameters and 4-bit quantization, its per-task cost is just 1/20 of Hy4 and 1/8 of DeepSeek-V4.1-Flash, with official claims of slightly better overall quality than Hy4 preview and plans to open-source.