Tencent's Hy Memory 2.0 Tested: Quality Depends on the LLM Writer
Tencent released Hy Memory 2.0, an Apache-2.0 framework with a 30B memory model and the MemHub app. An independent LongMemEval test found it scored 43% with Gemini 3.8 Flash versus 95% for Aura, but with Claude Opus 5.5 it matched Aura on knowledge updates (9/10) at about $5 per user history.
- Hy Memory 2.0 ships as an Apache-2.0 framework, 30B memory model and MemHub app
- With Gemini Flash, Hy scored 43% on LongMemEval versus 95% for Aura
- With Claude Opus 5.5, Hy matched Aura on knowledge updates at 9/10
- Writing memory via Opus costs about $5 per 50-conversation history versus $0 for Aura
Read next
AI