EVALUATION DIRECTORY

Benchmarks & Datasets

A consolidated repository of evaluations and environments referenced in the survey (Table 8). Filter by memory capabilities or environmental structures.

NameFac.Exp.M.M.EnvironmentScale / SizePrimary FeaturesLink
MemBenchsimulated53,000 samplesinteractive scenariosLink
MemoryAgentBenchsimulated4 tasksmulti-turn interactionsLink
LoCoMoreal300 samplesconversational memoryLink
WebChoreArenareal4 tasks / 532 samplestedious web browsingLink
MT-Mind2Webreal720 samplesconversational web navigationLink
PersonaMemsimulated15 tasks / 180 samplesdynamic user profilingLink
LongMemEvalsimulated5 tasks / 500 samplesinteractive memoryLink
PerLTQAsimulated8,593 samplessocial personalized interactionsLink
MemoryBanksimulated194 samplesuser memory updatingLink
MPRsimulated108,000 samplesuser personalizationLink
PrefEvalsimulated3,000 samplespersonal preferencesLink
LOCCOsimulated3,080 sampleschronological conversationsLink
StoryBenchmixed3 tasksinteractive fiction gamesLink
MemoryBenchsimulated4 tasks / ~20,000 samplescontinual learningLink
Madial-Benchsimulated331 samplesmemory recallingLink
Evo-Memorysimulated10 tasks / ~3,700 samplestest-time learningLink
LifelongAgentBenchsimulated1,396 sampleslifelong learningLink
StreamBenchsimulated9,702 samplescontinuous online learningLink
DialSimreal~1,300 samplesmulti-dialogue understandingLink
LongBenchmixed21 tasks / 4,750 sampleslong-context understandingLink
LongBench v2mixed20 tasks / 503 sampleslong-context multitasksLink
RULERsimulated13 taskslong-context retrievalLink
BABILongsimulated20 taskslong-context reasoningLink
MM-Needlesimulated~280,000 samplesmultimodal long-context retrievalLink
HaluMemsimulated3,467 samplesmemory hallucinationsLink
HotpotQAsimulated113k sampleslong-context QALink
ALFWorldsimulated3,353 taskstext-based embodied environmentLink
ScienceWorldsimulated10 t. / 30 t.interactive embodied environmentLink
AgentGymmixed89 tasks / 20,509 samplesmultiple environmentsLink
AgentBoardmixed9 tasks / 1013 samplesmulti-round interactionLink
WebShopsimulated12,087 samplese-commerce web interactionLink
WebArenareal812 samplesweb interactionLink
MMInAreal1,050 samplesmultihop web interactionLink
SWE-Bench Verifiedreal500 samplescode repairLink
GAIAreal466 sampleshuman-level deep researchLink
xBench-DSreal100 samplesdeep-search evaluationLink
ToolBenchreal126,486 samplesAPI tool useLink
GenAI-Benchreal~40,000 samplesvisual generation evaluationLink