Skip to content

Add ClawBench web-agent benchmark #6

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
reacher-z wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add ClawBench web-agent benchmark #6

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,7 @@ Save hours of research time—get straight to what matters.
| CompWoB | https://github.com/google-research/google-research/tree/master/compositional_rl/compwob | 2023 | Tests agents on sequential, compositional tasks that require state management across multiple steps, simulating real-world automation scenarios. | 50 compositional tasks | Task Success Rate | Element Match | Web |
| TURKING BENCH | https://turkingbench.github.io | 2024 | Uses natural HTML tasks from crowdsourcing to assess interaction skills with real-world web layouts and elements. | 32.2K instances | Task Success Rate | Text Match, Element Match, Image Match | Web |
| VisualWebBench | https://visualwebbench.github.io | 2024 | Provides a fine-grained assessment of multimodal large language models (MLLMs) on web-specific tasks. | 1,534 instances from 139 real websites across 87 sub-domains | Task Success Rate, Turn Success Rate, Efficiency Score, Metrics | Text Match, Image Match, Element Match, Action Match | Web |
| ClawBench | https://claw-bench.com | 2026 | Evaluates AI agents on everyday online tasks in isolated live-web environments with submission interception, execution traces, and judge-assisted outcome assessment. | 283 tasks across 163 live websites in V1+V2 | Task Success Rate, trajectory evidence, judge assessment | HTTP/request interception, browser actions, screenshots, recordings, agent messages | Web |
| WONDERBREAD | https://github.com/HazyResearch/wonderbread | 2024 | Focuses on business process management (BPM) tasks like documentation, knowledge transfer, and process improvement. | 2,928 human demonstrations across 598 distinct workflows | Task Success Rate, Step Success Rate, Efficiency Score, Completion under Policy | Text Match, Action Match, State Information | Web |
| WebOlympus | / | 2024 | An open platform for web agents that simplifies running demos, evaluations, and data collection for web agents on live websites. | 50 tasks | Task Success Rate, Step Success Rate | Action Match | Web |
| AndroidEnv | https://github.com/google-deepmind/android_env | 2021 | Provides an open-source platform based on the Android ecosystem with over 100 tasks across approximately 30 apps, focusing on reinforcement learning for various Android interactions. | 100+ tasks | NA | NA | Android |
Expand Down