Research map / Multi-agent systems and games / LLM agents
Mind2Web: Towards a Generalist Agent for the Web
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| multi-choice QA formulation instead of direct generation of target element (ablation) | MindAct Flan-T5-B · Mind2Web Cross-Task | step success rate: 17.5 -> 41.0 | Helped |
| multi-choice QA formulation instead of direct generation (ablation) | MindAct Flan-T5-B · Mind2Web Cross-Website | step success rate: 11.0 -> 29.5 | Helped |
| scale Flan-T5 from Base to XL (ablation) | MindAct · Mind2Web Cross-Task | step success rate: 41.0 -> 52.0 | Helped |
| scale Flan-T5 from Base to XL (ablation) | MindAct · Mind2Web Cross-Website | step success rate: 29.5 -> 38.9 | Helped |
| scale Flan-T5 from Base to XL (ablation) | MindAct · Mind2Web Cross-Domain | step success rate: 31.6 -> 39.6 | Helped |
| fine-tune Flan-T5-XL instead of zero-shot (ablation) | MindAct Flan-T5-XL · Mind2Web Cross-Task | element accuracy: 10.8 -> 52.0 | Helped |
| fine-tune Flan-T5-XL instead of zero-shot (ablation) | MindAct Flan-T5-XL · Mind2Web Cross-Website | element accuracy: 7.8 -> 38.9 | Helped |
| random grouping of candidate options (5 seeds) (ablation) | MindAct Flan-T5 · Mind2Web all splits | step success rate: std < 1 | No clear effect |
| test on unseen websites/domains instead of seen websites (cross-task) | MindAct Flan-T5-XL · Mind2Web | step success rate: 52.0 -> 38.9 (Cross-Website) | Hurt |
| use GPT-4 in-context (3 shots) instead of GPT-3.5 | MindAct · Mind2Web Cross-Website | element accuracy: 19.3 -> 35.8 | Helped |
Closest papers by meaning
- A Real-World WebAgent with Planning, Long Context Understanding, and Program Synthesis (ICLR 2024 · similarity 0.91)
- GPT-4V(ision) is a Generalist Web Agent, if Grounded (ICML 2024 · similarity 0.90)
- How to Train Your LLM Web Agent: A Statistical Diagnosis (NeurIPS 2025 · similarity 0.90)
- Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge (NeurIPS 2025 · similarity 0.90)
- AgentOccam: A Simple Yet Strong Baseline for LLM-Based Web Agents (ICLR 2025 · similarity 0.89)
- WebGym: Scaling Training Environments for Long-Horizon Visual Web Agents with Realistic Tasks (CVPR 2026 · 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.