Back to Arena

ExpeL

by Tsinghua University (Zhao et al.)

System Card

OrganizationTsinghua University (Zhao et al.)
Released2023-08
Architectureagentic-workflow / Natural-language insight extraction from training trajectories
DetailsAutonomously gathers experiences across training tasks, derives natural-language insights, and uses its own successful experiences as in-context examples at test time. No parameter updates — compatible with proprietary APIs.
Parameters—
Domainagent-memorylifelong-learning
Open SourceYes
PaperView Paper
CodeRepository
experiential-learninginsightsin-contextaaai-2024

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Data Transparency:1 self-reported5 estimated
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
2/2
LoCoMo
83.194pEstimated
MemoryBank
79.676pEstimated
Cross-Session Memory
1/1
LongMemEval
8186pEstimated
Multi-Hop QA
2/3
BABILong
71.212pEstimated
MultiHop-RAG
no data
HotpotQA
391pSelf-Reported
Agent Task Memory
1/1
AgentBench-Mem
7226pEstimated
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:arXiv:2308.10144 Figure 5 — Success rate read from Figure 5; not a precise table cellArena estimate — derived from capability profile, not independently verified