Tencent has released Hy Memory 2.0, an open-source stack designed to manage what AI agents remember. The system includes a 30B-parameter model specialized for memory writing, a Python SDK that categorizes memory into chat, working, and coding types, and a local-first app called MemHub for cross-tool history synchronization. While Tencent claims parity with Claude Opus on memory benchmarks, these results are self-reported. The Hy Memory 2.0 stack aims to improve AI agent efficiency by offloading the memory management task to a dedicated, smaller model. AI
IMPACT Specialized memory models could improve AI agent efficiency and reduce token usage for frontier models.
RANK_REASON Release of an open-source model and framework for a specialized AI task. [lever_c_demoted from research: ic=1 ai=1.0]
- Apache Software License 2.0
- Claude Opus
- DeepSeek
- Hugging Face
- Hy Memory 2.0
- Hy-Memory-30B-A3B
- Mem0 Agent Memory Framework
- MemHub
- Meta*
- Muse Wonder
- NeurAITool
- Tencent
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →