Unverified50% confidenceBenchmarkExact time
MemoryBench is a benchmark created by Meta itself, testing recall of key information in multi-turn long conversations of 64K to 100K tokens, with conversations spanning multiple facts and including distractor conversations as well as injected turns that conflict with or update the target details, scored by an LLM judge according to weighted scoring criteria.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/15/2026
First Seen
Valid until: 1/13/2027
Sources
Related Entities
Related Claims
VerifiedMulti-turn conversation configuration addresses the context memory problem, determining how many rounds of historical information an agent can retain during continuous dialogue.65% similarUnverifiedDue to the combination of multiple rounds of iteration, full context carrying, and self-checking mechanisms, token consumption in Loop Engineering is typically 5 to 20 times that of ordinary conversations.64% similarUnverifiedLongMemEval is a benchmark for evaluating long-horizon memory capability, in which multi-hop tasks require the system to reason across multiple memory fragments to answer a single question.64% similarUnverifiedMeta describes two model features: Active Memory carries relevant context across conversations, and Active Focus keeps trying different paths on difficult goals and confirms before critical actions.63% similarUnverifiedMemoryOps's core working loop consists of four steps: RECALL, REFLECT, RECOMMEND, and RETAIN.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1034408API
curl https://kongchang.com/api/v1/knowledge/claims/1034408MCP
get_claim(id=1034408)