AI memory startup MemoraX AI has closed a Seed++ round worth hundreds of millions of yuan, led jointly by Shangshi Capital, Haitang Capital, and Renai Capital, with follow-on participation from other institutions and existing backers.
The company was founded in March and has now raised three rounds in a row — in April, May, and August.
Founder Hao Jianye previously led Huawei's Decision Intelligence Lab and Large Model Algorithm Lab, and served as technology president for Huawei's healthcare business unit. He's also an elite professor at Tianjin University. MemoraX's first round brought in over $10 million, its second was tens of millions of yuan, and the first two rounds combined already exceeded 100 million yuan.
MemoraX's focus is long-term memory for large language models. Rather than relying only on vector databases to store and retrieve historical context, the company is training dedicated memory models via reinforcement learning — teaching AI what's worth remembering and when to recall it.
Its memory system previously scored 82.65 on LoCoMo-Refined, and the company also helped build two long-term memory benchmarks: ScriptMem and SWE-context-bench.
On SWE-ContextBench Lite, MemoraX posted a 45.45% task completion rate, compared with 29.20% for OpenViking — a 16.25-percentage-point lead. Supermemory and Mem0 landed at 30.30% and 24.24%, respectively.
The company has also launched MemoraX Code, which provides cross-session long-term memory for Claude Code and Codex. It saves codebase knowledge, engineering experience, and user habits, letting coding agents pick up where they left off after opening a new session.
Via ChinaVenture