Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

MemoryAgentBench

Benchmarks

Task NameDataset NameSOTA ResultTrend
Memory Agent PerformanceMemoryAgentBench
Access Rate83
44
Memory-Augmented Agent EvaluationMemoryAgentBench
Single-Hop Accuracy28.7
16
Long-context memory managementMemoryAgentBench
Single-Doc Recall83
16
End-to-end agent-loop performanceMemoryAgentBench
Pre-Verify Success Rate80
10
Conflict ResolutionMemoryAgentBench conflict-bearing subset official
Final Accuracy100
10
Agentic Memory RetrievalMemoryAgentBench
Access Rate82
10
Fact Consolidation (Single-Hop)MemoryAgentBench (MAB) FC-SH 262K
Accuracy93
8
Conflict resolution and memory verificationMemoryAgentBench conflict-bearing slice
Accuracy100
7
Fact Consolidation (Multi-Hop)MemoryAgentBench (MAB) FC-MH 262K
Accuracy27
5
Fact Consolidation (Single-Hop)MemoryAgentBench FC-SH Average
Accuracy0.948
4
Fact Consolidation (Single-Hop)MemoryAgentBench (MAB) FC-SH 64K
Accuracy95
4
Fact Consolidation (Single-Hop)MemoryAgentBench (MAB) FC-SH 32K
Accuracy92
4
Fact Consolidation (Multi-Hop)MemoryAgentBench (MAB) FC-MH Average
Accuracy30.2
1
Fact Consolidation (Multi-Hop)MemoryAgentBench (MAB) FC-MH 64K
Accuracy33
1
Fact Consolidation (Multi-Hop)MemoryAgentBench FC-MH 32K
Accuracy27
1
Fact Consolidation (Multi-Hop)MemoryAgentBench (MAB) FC-MH 6K
Accuracy34
1
Showing 16 of 16 rows