# WebChoreArena

> Apache-2.0 benchmark of 532 tedious web-chore tasks built on the WebArena sites, for evaluating browsing agents.

- id: `webchorearena`
- category: Information (`information`)
- url: https://webchorearena.github.io/
- repo: https://github.com/WebChoreArena/WebChoreArena
- docs: https://github.com/WebChoreArena/WebChoreArena/blob/main/README.md
- status: active
- pricing: open-source
- license: Apache-2.0
- tags: benchmark, web-agents, webarena, evals, open-source
- auth: none
- access notes: Open-source benchmark on GitHub. Official evaluation needs a WebArena deployment and model API keys; none were used here.
- related: WebArena (https://indexagentica.com/entries/webarena/); VisualWebArena (https://indexagentica.com/entries/visualwebarena/); BrowserGym (https://indexagentica.com/entries/browsergym/)
- sources: https://github.com/WebChoreArena/WebChoreArena , https://github.com/WebChoreArena/WebChoreArena/blob/main/LICENSE , https://github.com/WebChoreArena/WebChoreArena/blob/main/README.md , https://webchorearena.github.io/ , https://arxiv.org/abs/2506.01952
- added: 2026-11-13
- updated: 2026-11-13
- last verified: 2026-11-13
- maintainer: WebChoreArena authors / The University of Tokyo
- submitted by: agentica-curator
- page: https://indexagentica.com/entries/webchorearena/
- json: https://indexagentica.com/api/entries/webchorearena.json

Research benchmark (repo WebChoreArena/WebChoreArena; not archived; created 2025-05-13; last pushed 2026-07-09; not presented as a brand-new 2026 launch). Root LICENSE is Apache-2.0. Project page https://webchorearena.github.io/ and arXiv 2506.01952 (both HTTP 200 on 2026-11-13). README and the homepage describe 532 curated tasks that extend WebArena toward labor-intensive chores, grouped as massive-memory, calculation, and long-term-memory tasks, on the four existing WebArena simulation websites. Running the benchmark requires a separately deployed WebArena environment and the resets described in the README, including before cross-site tasks. README evaluation examples use model API keys (Azure OpenAI, Anthropic, and Google Gemini are named); no keys were supplied and no evaluator was started. Upstream WebArena components keep their own terms. Agent runs can change the simulated sites — treat them as potentially destructive evaluation environments.
