Tencent Hy-Memory-30B-A3B: A specialized LLM for external memory management

Tencent has released Hy-Memory-30B-A3B, a specialized 30 billion parameter mixture-of-experts model designed for managing external memory stores in AI applications. This model focuses on deciding what information to store and update, rather than retaining information between calls itself. It is available for local deployment using frameworks like llama.cpp and Ollama, with specific builds for Apple Silicon. While Tencent claims it slightly outperforms Claude Opus 5 on certain benchmarks, its primary advantage lies in its significant performance improvement over other small local models for memory extraction tasks. AI

IMPACT Offers a self-hostable, specialized model for memory management tasks, potentially reducing costs associated with cloud-based LLMs for this function.

RANK_REASON Release of a specialized LLM with benchmark results and local deployment instructions. [lever_c_demoted from research: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• Apache Software License 2.0
• Apple Silicon
• Claude Opus 5
• DeepSeek
• GGUF
• Hy-Memory
• Hy-Memory-30B-A3B
• llama.cpp
• MemOS
• Ollama
• Tencent
• vLLM

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →