REVIEW 19 cited by
WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Large language models (LLMs) have shown remarkable potential as autonomous agents, particularly in web-based tasks. However, existing LLM web agents heavily rely on expensive proprietary LLM APIs, while open LLMs lack the necessary decision-making capabilities. This paper introduces WebRL, a self-evolving online curriculum reinforcement learning framework designed to train high-performance web agents using open LLMs. WebRL addresses three key challenges in building LLM web agents, including the scarcity of training tasks, sparse feedback signals, and policy distribution drift in online learning. Specifically, WebRL incorporates 1) a self-evolving curriculum that generates new tasks from unsuccessful attempts, 2) a robust outcome-supervised reward model (ORM), and 3) adaptive reinforcement learning strategies to ensure consistent improvements. We apply WebRL to transform open Llama-3.1 and GLM-4 models into proficient web agents. On WebArena-Lite, WebRL improves the success rate of Llama-3.1-8B from 4.8% to 42.4%, and from 6.1% to 43% for GLM-4-9B. These open models significantly surpass the performance of GPT-4-Turbo (17.6%) and GPT-4o (13.9%) and outperform previous state-of-the-art web agents trained on open LLMs (AutoWebGLM, 18.2%). Our findings demonstrate WebRL's effectiveness in bridging the gap between open and proprietary LLM-based web agents, paving the way for more accessible and powerful autonomous web interaction systems.
Forward citations
Cited by 19 Pith papers
-
Progressive Agent Skill Generation via Reinforcement Learning
Skill-α uses reinforcement learning with a rollback reward to progressively edit SKILL.md-style procedure files, improving downstream success rates over pipeline baselines on CL-Bench, SpreadsheetBench, and tau2-bench.
-
OSReward: Instituting Standardized Evaluation for Cross-Platform Computer-Use Reward Models
VLM judges of CUA trajectories are systematically lenient; OSReward measures this with human gold, and OS-Shepherd open models close most of the reliability gap cheaply.
-
SCALECUA: Scaling Computer Use Agents with Verifiable Task Synthesis and Efficient Online RL
Verifiable task synthesis plus frontier sampling and sliding-window visual training yields open-source SOTA computer-use agents on OSWorld and ScienceBoard.
-
Agentic-DPO: From Imitation to Agentic Policy Optimization on Expert Trajectories
Offline DPO on expert states with one-step student negatives and schema-preserving augmentation improves LLM agents beyond SFT and can match online GRPO without environment interaction.
-
SOD: Step-wise On-policy Distillation for Small Language Model Agents
SOD reweights on-policy distillation strength step-by-step using divergence to stabilize tool use in small language model agents, yielding up to 20.86% gains and 26.13% on AIME 2025 for a 0.6B model.
-
GUI-Libra: Training Native GUI Agents to Reason and Act with Action-aware Supervision and Partially Verifiable RL
Action-weighted SFT plus KL-regularized GRPO on a curated 81K reasoning dataset lifts open-source GUI agents by 11–22 points on online task-completion benchmarks.
-
IPR-1: Interactive Physical Reasoner
IPR uses world-model rollouts to reinforce a VLM policy via PhysCode on a 1000+ game benchmark, achieving robust physical reasoning that improves with experience and transfers zero-shot to unseen games while surpassing GPT-5.
-
Agent Learning via Early Experience
Training agents to predict consequences of their own alternative actions (implicit world modeling) or to reflect on why expert actions were better (self-reflection) consistently improves task success, OOD generalizati...
-
A global log for medical AI
MedLog defines a nine-field, syslog-style event log for clinical AI, intended to support real-world surveillance and auditing; the four-deployment validation claimed in the abstract is absent from the body.
-
Symbolic Graphics Programming with Large Language Models
Qwen-2.5-7B trained with reinforcement learning against SigLIP visual similarity scores writes SVG drawings that match text captions about as well as frontier models on several automated metrics.
-
CODA: Coordinating the Cerebrum and Cerebellum for a Dual-Brain Computer Use Agent with Decoupled Reinforcement Learning
A decoupled planner-executor GUI agent, trained by per-app reinforcement learning followed by specialist-to-generalist distillation, lifts ScienceBoard success from about 7.6% to 21.0% average and 40% pass@8.
-
SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience
A self-evolving computer-use agent trained with full-trajectory state judging and curriculum task generation goes from 11.3% to 34.5% average success on five OSWorld apps, and a specialist-to-generalist variant beats ...
-
WebSynthesis: World-Model-Guided MCTS for Efficient WebUI-Trajectory Synthesis
A world-model-guided MCTS pipeline synthesizes 4k web navigation trajectories and yields a WebArena Pass@3 success rate of 20.15%, above OS-Genesis (18.66%) and AgentTrek (11.94%).
-
Agent-RLVR: Training Software Engineering Agents via Guidance and Environment Rewards
A guidance-augmented RLVR framework improves software engineering agent pass@1 on SWE-Bench Verified from 9.4% to 22.4%, with guidance being the critical component.
-
Build the web for agents, not agents for the web
The paper proposes a paradigm shift: design a standardized Agentic Web Interface for AI agents, rather than adapting agents to human-facing websites.
-
Atom-Searcher: Enhancing Agentic Deep Research via Fine-Grained Atomic Thought Reward
A new RL framework that rewards fine-grained reasoning steps, called Atomic Thoughts, claims better agentic deep research on seven benchmarks.
-
EvoCurr: Self-evolving Curriculum with Behavior Code Generation for Complex Decision-making
EvoCurr couples an LLM curriculum designer with an LLM code-generating solver, but its only reported success is 1 of 5 runs and no direct baseline is shown.
-
Cognitive Duality for Adaptive Web Agents
A web agent called CogniWeb reportedly reaches 43.96% success on WebArena with 75% fewer tokens by switching between fast and slow reasoning, but the supplied body is a different paper and the claims could not be checked.
-
Atomic-to-Compositional Generalization for Mobile Agents with A New Benchmark and Scheduling System
AI phone agents succeed on single-step commands but fail on multi-step compositions; a new benchmark measures this gap and a scheduler that decomposes tasks recovers most of it.
Discussion (0). Sign in to comment.