Back to Arena

WebVoyager

by MinorJerry et al.

System Card

OrganizationMinorJerry et al.
Released2024-01
Architectureagentic-workflow / Multimodal web agent with action/state history
DetailsLMM-powered web agent on Selenium that navigates real websites, uses screenshots + DOM, and maintains trajectory memory across 643 test tasks plus GAIA tasks.
Parameters—
Domainagent-memory
Open SourceYes
PaperView Paper
CodeRepository
lmmseleniumgaiaweb-benchmark

Capability Profile

Benchmark Scores

5 of 14 benchmarks
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
1/2
LoCoMo
76.664p
MemoryBank
no data
Cross-Session Memory
1/1
LongMemEval
79.576p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
74.270p
HotpotQA
77.788p
Agent Task Memory
1/1
AgentBench-Mem
7226p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:WebVoyager paper (arXiv:2401.13919); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)WebVoyager paper (arXiv:2401.13919); evaluated on HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering (Stanford / CMU, 1809)WebVoyager paper (arXiv:2401.13919); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)WebVoyager paper (arXiv:2401.13919); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)WebVoyager paper (arXiv:2401.13919); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)