Tencent releases Hy-Memory-30B-A3B, a memory writer model for long-term AI memory

Tencent Hunyuan releases Hy-Memory-30B-A3B, a compact writer model with about 3B active parameters that decides what to retain in external memory frameworks such as Mem0, MemOS, and Hy-Memory.

The model is trained in two stages, Memory On-Policy Distillation followed by Attributed Memory Reinforcement Learning.

Across four harnesses on a fixed 20% benchmark subsample, its overall average is 55.49, ahead of Claude Opus 5 at 54.62 and DeepSeek V4.1 Flash at 53.64. The model card provides weights, a vLLM deployment example, and a technical report.