List your company

Long-term Memory Evaluation on LongMemEvalS

95.6Overall ScoreClaude Opus 4.6
34.86450.63266.482.168Mar 17, 2026Mar 22, 2026Mar 27, 2026Apr 2, 2026Apr 7, 2026Apr 12, 2026Apr 18, 2026

Evaluation results

MethodLinks
Claude Opus 4.62026.0395.610088.7210010098.5795.49
GPT-5-mini2026.0394.296.1585.7198.2110097.1496.99
Claude Sonnet 4.52026.0394.298.7289.4710093.3398.5791.73
Claude Haiku 4.52026.039496.1588.721009097.1494.74
GPT-5.22026.0393.897.4483.4610010098.5795.49
MemMachineModel=gpt-5-mini, Ablation setting=Ablation best2026.0493——————
GPT-4o2026.0392.696.1591.731008094.2990.23
Claude Code Sonnet2026.0388.693.5979.794.648094.2990.98
MemCoTModel=GPT-4o-mini [4]2026.048888.479.61007098.589.4
MnemisModel=GPT-4o-mini [4]2026.0487.292.376.71009097.183.5
EMem-GModel=GPT-4o-mini [4]2026.0477.994.473.687.532.28774.8
Mem0Model=GPT-4o-mini [4]2026.0471.174.466.296.43491.463.9
RAGModel=GPT-4o-mini [4]2026.0467.270.547.491.17088.663.2
HeLa-MemBackbone=GPT-4o-mini, learning rate η=0.02, decay rate λ=0.995, spreading activation strength β=0.1, spreading threshold θ=0.4, keyword weight=0.7, max flipped items m=32026.0465.4——————
NemoriModel=GPT-4o-mini [4]2026.0464.261.551.183.946.788.661.7
ZepModel=GPT-4o-mini [4]2026.0463.274.447.47553.392.954.1
A-MEMBackbone=GPT-4o-mini2026.0462.6——————
NaiveRAGBackbone=GPT-4o-mini2026.0461——————
FullTextBackbone=GPT-4o-mini2026.0456.8——————
Full ContextModel=GPT-4o-mini [4]2026.045578.238.389.36.778.642.1
Mem0Backbone=GPT-4o-mini2026.0453.61——————
MemoryOSBackbone=GPT-4o-mini2026.0444.8——————
LangMemBackbone=GPT-4o-mini2026.0437.2——————