EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments
Pretraining scaling laws reveal that model capability improves predictably with data and compute. But learning from real world environments after deployment remains far less understood. Analyzing roughly 38,000 hours of agent interaction with the environment across 134 real world tasks, we find, to the best of our knowledge, the first evidence that overall performance during environment learning follows a log-sigmoid scaling law with remarkably high precision, reaching R^2 = 0.998. Across model generations, we also find that agent learning speed roughly doubles every three months. This discove
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 55%Alibaba's model never trained as an agent — and improved agent performance across seven benchmarks →
- PossiblePossibly related (embedding) · 54%TIGER-AI-Lab/ClawBench →
- PossiblePossibly related (embedding) · 51%AgentToolkit/altk-evolve →
- PossiblePossibly related (embedding) · 51%Unity-Technologies/ml-agents →
- PossiblePossibly related (embedding) · 51%xlang-ai/OSWorld →
- LinkedLinked via arxiv author · 85%Deyao Zhu →
“EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments”
- LinkedLinked via arxiv author · 85%Xin Zhou →
“EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments”
- LinkedLinked via arxiv author · 85%Shengling Qin →
“EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments”