Agentic Test-Time Scaling for WebAgents
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908830310334464 |
|---|---|
| author | Lee, Nicholas Erdogan, Lutfi Eren John, Chris Joseph Krishnapillai, Surya Mahoney, Michael W. Keutzer, Kurt Gholami, Amir |
| author_facet | Lee, Nicholas Erdogan, Lutfi Eren John, Chris Joseph Krishnapillai, Surya Mahoney, Michael W. Keutzer, Kurt Gholami, Amir |
| contents | Test-time scaling has become a standard way to improve performance and boost reliability of neural network models. However, its behavior on agentic, multi-step tasks remains less well-understood: small per-step errors can compound over long horizons; and we find that naive policies that uniformly increase sampling show diminishing returns. In this work, we present CATTS, a simple technique for dynamically allocating compute for multi-step agents. We first conduct an empirical study of inference-time scaling for web agents. We find that uniformly increasing per-step compute quickly saturates in long-horizon environments. We then investigate stronger aggregation strategies, including an LLM-based Arbiter that can outperform naive voting, but that can overrule high-consensus decisions. We show that uncertainty statistics derived from the agent's own vote distribution (entropy and top-1/top-2 margin) correlate with downstream success and provide a practical signal for dynamic compute allocation. Based on these findings, we introduce Confidence-Aware Test-Time Scaling (CATTS), which uses vote-derived uncertainty to allocate compute only when decisions are genuinely contentious. CATTS improves performance on WebArena-Lite and GoBrowse by up to 9.1% over React while using up to 2.3x fewer tokens than uniform scaling, providing both efficiency gains and an interpretable decision rule. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_12276 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Agentic Test-Time Scaling for WebAgents Lee, Nicholas Erdogan, Lutfi Eren John, Chris Joseph Krishnapillai, Surya Mahoney, Michael W. Keutzer, Kurt Gholami, Amir Artificial Intelligence Computation and Language Test-time scaling has become a standard way to improve performance and boost reliability of neural network models. However, its behavior on agentic, multi-step tasks remains less well-understood: small per-step errors can compound over long horizons; and we find that naive policies that uniformly increase sampling show diminishing returns. In this work, we present CATTS, a simple technique for dynamically allocating compute for multi-step agents. We first conduct an empirical study of inference-time scaling for web agents. We find that uniformly increasing per-step compute quickly saturates in long-horizon environments. We then investigate stronger aggregation strategies, including an LLM-based Arbiter that can outperform naive voting, but that can overrule high-consensus decisions. We show that uncertainty statistics derived from the agent's own vote distribution (entropy and top-1/top-2 margin) correlate with downstream success and provide a practical signal for dynamic compute allocation. Based on these findings, we introduce Confidence-Aware Test-Time Scaling (CATTS), which uses vote-derived uncertainty to allocate compute only when decisions are genuinely contentious. CATTS improves performance on WebArena-Lite and GoBrowse by up to 9.1% over React while using up to 2.3x fewer tokens than uniform scaling, providing both efficiency gains and an interpretable decision rule. |
| title | Agentic Test-Time Scaling for WebAgents |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2602.12276 |