Research map / Multi-agent systems and games / LLM agents
MolmoWeb: Open Visual Web Agent and Open Data for the Open Web
Tanmay Gupta, Piper Wolters, Zixian Ma, Peter Sushko, Rock Yuren Pang, Yue Yang, Jason Ren, Harsh Trivedi, Taira Anderson, Winson Han, Ranjay Krishna
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| scale training data from 1% to 10% (ablation) | MolmoWeb · WebVoyager (30 steps) | success rate (%): 44.5 -> 63.2 | Improved |
| scale training data from 10% to 100% (ablation) | MolmoWeb · WebVoyager (30 steps) | success rate (%): 63.2 -> 68.5 | Improved |
| add human trajectories to synthetic trajectories (ablation) | MolmoWeb · WebVoyager (30 steps) | success rate (%): 67.8 -> 68.5 | No clear effect |
| add human trajectories to synthetic trajectories (ablation) | MolmoWeb · Online-Mind2Web (30 steps) | success rate (%): 22.0 -> 21.4 | No clear effect |
| use synthetic AxTree-agent trajectories instead of human ones (same tasks, 2.7K) (ablation) | MolmoWeb · WebVoyager | success rate (%): 35.4 -> 53.0 | Improved |
| use synthetic instead of human trajectories (2.7K) (ablation) | MolmoWeb · Online-Mind2Web | success rate (%): 9.0 -> 16.8 | Improved |
| use synthetic instead of human trajectories (2.7K) (ablation) | MolmoWeb · DeepShop | success rate (%): 19.8 -> 24.4 | Improved |
| sample with top-p instead of greedy decoding (ablation) | MolmoWeb · WebVoyager | success rate (%): 61.4 -> 68.5 | Improved |
| sample with top-p (p=0.8) instead of top-k (k=20) (ablation) | MolmoWeb · WebVoyager | success rate (%): 67.4 -> 68.5 | Improved |
| use planner-operator-verifier multi-agent harness instead of Gemini AxTree agent alone for trajectory generation (ablation) | Gemini AxTree agent · WebVoyager | success rate (%): 74.4 -> 78.5 | Improved |
| use parallel rollouts with best-of-4 selection instead of single rollout | MolmoWeb-8B · WebVoyager | success rate (%): 78.2 -> 94.7 | Improved |
| use parallel rollouts with best-of-4 selection instead of single rollout | MolmoWeb-8B · Online-Mind2Web | success rate (%): 35.3 -> 60.5 | Improved |
Closest papers by meaning
- AgentOccam: A Simple Yet Strong Baseline for LLM-Based Web Agents (ICLR 2025 · similarity 0.90)
- WebGym: Scaling Training Environments for Long-Horizon Visual Web Agents with Realistic Tasks (CVPR 2026 · similarity 0.89)
- A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis (ICLR 2024 · similarity 0.89)
- AgentWebBench: Benchmarking Multi-Agent Coordination in Agentic Web (ICML 2026 · similarity 0.89)
- REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites (NeurIPS 2025 · similarity 0.89)
- How to Train Your LLM Web Agent: A Statistical Diagnosis (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
