Web Navigation and Task Automation on WebArena Continual-memory deployment
52.3Success Rate (SR)Auto-Dreamer
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Auto-DreamerMemory Category=Ours, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview, Consolidator LLM=Qwen3-14B (SciWorld-trained)2026.05 | 52.3 | 927 | 0.628 | |
| AWMMemory Category=Workflow / procedural memory, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 52 | 890 | 0.597 | |
| LightMemMemory Category=Two-timescale prompted, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 52 | 370,874 | 0.567 | |
| ExpeLMemory Category=Reflective / insight extraction, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 50.8 | 3,371 | 0.576 | |
| MempMemory Category=Workflow / procedural memory, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 50.8 | 4,973 | 0.574 | |
| Mem0Memory Category=Structured stores, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 50.3 | 43,355 | 0.571 | |
| ReasoningBankMemory Category=Structured stores, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 49.8 | 24,362 | 0.581 | |
| ReflexionMemory Category=Reflective / insight extraction, Frozen Task Agent LLM=Gemini-3-flash-preview, Memory-generation LLM=Gemini-3.1-flash-lite-preview2026.05 | 46.4 | 8,148 | 0.567 | |
| No memoryMemory Category=Baseline, Frozen Task Agent LLM=Gemini-3-flash-preview2026.05 | 44.6 | 0 | 0.527 |