Tencent releases Hy-Memory-30B-A3B open-source memory writer model and Hy Memory 2.0

Tencent Hunyuan releases Hy-Memory-30B-A3B, a memory 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 and then Attributed Memory Reinforcement Learning. On a fixed 20% benchmark subsample across four harnesses, Tencent reports an overall average of 55.49, ahead of Claude Opus 5 at 54.62 and DeepSeek V4.1 Flash at 53.64. The model card on Hugging Face provides the weights, a vLLM deployment example, and a technical report.

Tencent Hunyuan later announces Hy Memory 2.0, which includes the open-source model, a framework that handles chat, work, and code memories, and MemHub, a CLI and Mac tool. Tencent says Hy-Memory-30B-A3B is #1 open-source on 4 benchmarks and beats Opus 5 across major frameworks, and that MemHub syncs memories across 10+ AI tools. These are Tencent's claims in its announcement post; the benchmark results come from its own evaluation.