# OSU-NLP-Group/Mind2Web

[NeurIPS'23 Spotlight] "Mind2Web: Towards a Generalist Agent for the Web" -- the first LLM-based web agent and benchmark for generalist web agents

Repository: https://github.com/OSU-NLP-Group/Mind2Web
Canonical: https://ross.abutalabs.com/products/mind2web
Homepage: https://osu-nlp-group.github.io/Mind2Web/
Language: Jupyter Notebook
License: MIT
License Family: permissive
Last push: 2025-11-05T00:38:41+00:00

## Health v2 (maintenance only)
Score: 52/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 50, release rhythm 35, longevity 87
- inputs: {"age_days": 1230, "days_push": 302, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1021, forks 124 (observed 2026-08-28T04:03:15.800305+00:00)

## What it is
Mind2Web is the first dataset and benchmark for developing and evaluating generalist LLM-based web agents that follow natural language instructions to complete complex tasks on real-world websites. It contains over 2,000 open-ended tasks from 137 websites across 31 domains, with crowdsourced action sequences, plus fine-tuning code and models for element ranking and action prediction.

## Use cases
- evaluate generalist web agents on real-world websites
- train LLM-based web navigation agents
- benchmark element ranking and action prediction on web pages
- research cross-domain generalization of web agents
- build agents that follow language instructions on any website
- compare web agent performance with macro and micro accuracy metrics

## When to choose
- you need a large-scale, diverse benchmark for web agents using real websites rather than simulated ones
- you want to train or fine-tune models for web task automation from language instructions
- you need crowdsourced human action sequences as supervision for web navigation research
- you want paired HTML and screenshot data for multimodal web agent research

## When to avoid
- you need a live interactive online evaluation environment rather than offline traces (see Online-Mind2Web or SeeAct instead)
- you only target a small set of specific websites or a simulated sandbox environment
- you need production-ready web automation rather than research benchmarking
- you cannot access the encrypted test set or agree to data contamination prevention terms

## Facets
- artifact type: dataset
- maturity: active
- function: machine-learning, agent-framework, rag, data-science, benchmarking, nlp
- domain: artificial-intelligence, large-language-models, machine-learning, web-development, data-science
- platform: python, cross-platform
- tags: web-agent, benchmark, llm-agent, dataset, neurips-2023, web-navigation, html-understanding, huggingface, jupyter-notebook, academic-research, task-automation, element-ranking, action-prediction, ai-agents, natural-language-processing, research, web-server, gpu

## Member repositories
- OSU-NLP-Group/Mind2Web (main) score 52

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:15.800305+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:08:48.449807+00:00, confidence not recorded.
  - readme: https://github.com/OSU-NLP-Group/Mind2Web (fetched 2026-08-28T04:03:15.800305+00:00, sha 2fc60ba492b0)
  - homepage: https://osu-nlp-group.github.io/Mind2Web/ (fetched 2026-08-29T13:09:10.837575+00:00, sha 9fe2609f9ee9)
- Data as of 2026-08-30T08:39:29.467469+00:00.