From 52b69feca4a422ecf071610c06a29a39640dffc3 Mon Sep 17 00:00:00 2001 From: reacher-z Date: Mon, 27 Jul 2026 17:49:47 +0800 Subject: [PATCH] docs: reference ClawBench as an external benchmark --- README.md | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/README.md b/README.md index a19a6777..b31f75b3 100644 --- a/README.md +++ b/README.md @@ -22,6 +22,12 @@ with randomly-generated parameters to create millions of unique task variations. In addition to the built-in tasks, AndroidWorld also supports the popular web benchmark, MiniWoB++ from [Liu et al.](http://arxiv.org/abs/1802.08802). +### Related external benchmarks + +The following independent benchmark is related to AndroidWorld's autonomous-agent evaluation scope but is not integrated with or supported by AndroidWorld: + +* **[ClawBench](https://github.com/TIGER-AI-Lab/ClawBench)** — [paper](https://arxiv.org/abs/2604.08523) and [project page](https://claw-bench.com/). It evaluates web and computer-use agents on 283 tasks across 163 live platforms with isolated browser sessions, request-level outcome checks, and recorded action, screenshot, network, and message traces. See the project's own [runner and setup instructions](https://github.com/TIGER-AI-Lab/ClawBench#readme); task definitions are available on [Hugging Face](https://huggingface.co/datasets/NAIL-Group/ClawBench) and mirrored at [TIGER-Lab/ClawBench](https://huggingface.co/datasets/TIGER-Lab/ClawBench). + Key features of AndroidWorld include: * 📝 **116 diverse tasks** across 20 real-world apps