← All systems
Mem0
OrganizationMem0 Inc. (YC)
Founded2023 (as Embedchain)
Funding$24M — $20M Series A (Basis Set) + seed (Kindred), 2025-10
CodeApache-2.0 · 61.8k★
Two-phase LLM extraction/update into a vector store; optional graph variant (Mem0ᵍ).
Every number on file
| claimed | benchmark | config | runner | status |
|---|---|---|---|---|
| 66.88 / 68.44ᵍ | LoCoMo (J) | GPT-4o-mini · judge unnamed · adversarial excluded | self (paper) | SELF-REPORTED |
| 92.5 | LoCoMo | undisclosed · 2026 algorithm | self (blog) | SELF-REPORTED |
| 94.4 | LongMemEval | undisclosed · managed platform | self (blog) | SELF-REPORTED |
| 61.69 → 39.51 | LoCoMo/LME ACC | GPT-4o-mini judge, by reader | LightMem (ICLR 2026) | THIRD-PARTY |
| 21.1 (< BM25 41.5) | MemoryAgentBench | independent harness | UCSD (ICLR 2026) | THIRD-PARTY |
| 11.62 F1 | LoCoMo F1 | GPT-4o-mini, single run | O-Mem paper | THIRD-PARTY |
Own paper's full-context baseline (72.90) sits above its own score. README warns the OSS SDK won't match managed-platform numbers.
Rows are not comparable — configs differ. A row graduates to a verdict only through a registered re-run. The protocol →