# How to Train Your LLM Web Agent: A Statistical Diagnosis Source: [How to Train Your LLM Web Agent: A Statistical Diagnosis]() Collection index: [llms.txt]() Data updated: 2026-02-17T05:00:14.939Z Authors: Dheeraj Vattikonda, Santhoshi Ravichandran, Emiliano Penaloza, Hadi Nekoei, Megh Thakkar, Thibault Le Sellier de Chezelles, Nicolas Gontier, Miguel Mu\\~noz-M\\'armol, Sahar Omidi Shayegan, Stefania Raimondo, Xue Liu, Alexandre Drouin, Laurent Charlin, Alexandre Pich\\'e, Alexandre Lacoste, Massimo Caccia Published: 2025-07-05T17:12:33.000Z ## Abstract LLM-based web agents have recently made significant progress, but much of it has occurred in closed-source systems, widening the gap with open-source alternatives. Progress has been held back by two key challenges: first, a narrow focus on single-step tasks that overlooks the complexity of multi-step web interactions; and second, the high compute costs required to post-train LLM-based web agents. To address this, we present the first statistically grounded study on compute allocation for LLM web-agent post-training. Our approach uses a two-stage pipeline, training a Llama 3.1 8B student to imitate a Llama 3.3 70B teacher via supervised fine-tuning (SFT), followed by on-policy reinforcement learning. We find this process highly sensitive to hyperparameter choices, making exhaustive sweeps impractical. To spare others from expensive trial-and-error, we sample 1,370 configurations and use bootstrapping to estimate effective hyperparameters. Our results show that combining SFT with on-policy RL consistently outperforms either approach alone on both WorkArena and MiniWob++. Further, this strategy requires only 55% of the compute to match the peak performance of pure SFT on MiniWob++, effectively pushing the compute-performance Pareto frontier, and is the only strategy that can close the gap with closed-source models. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Web Task Completion on MiniWoB++ Held-out Goals (test)](): Web Task Completion; MiniWoB++ Held-out Goals (test); Success Rate: 67.2. - [Web Navigation and Automation on WorkArena Held-out Goals (test)](): Web Navigation and Automation; WorkArena Held-out Goals (test); Success Rate: 35.4. - [Web Navigation and Automation on WorkArena Held-out Tasks (test)](): Web Navigation and Automation; WorkArena Held-out Tasks (test); Success Rate: 28.8. - [Web Task Completion on MiniWoB++ Held-out Tasks (test)](): Web Task Completion; MiniWoB++ Held-out Tasks (test); Success Rate: 63.3. ## Original sources and implementation [Original paper]() [PDF]()