web-arena-x/visualwebarena
A realistic benchmark for evaluating multimodal autonomous language agents on diverse web-based visual tasks.
Not currently ranked — collecting fresh signals.
star history
VisualWebArena is a benchmark framework for evaluating multimodal agents through diverse and complex web-based visual tasks. It builds on WebArena with a reproducible, execution-based evaluation approach. The benchmark includes agent trajectories (such as GPT-4V with Set-of-Marks) and provides tools for end-to-end training and environment reset, enabling systematic assessment of multimodal agent capabilities.
Frequently asked
- What is web-arena-x/visualwebarena?
- A realistic benchmark for evaluating multimodal autonomous language agents on diverse web-based visual tasks.
- Is visualwebarena open source?
- Yes — web-arena-x/visualwebarena is open source, released under the MIT license.
- What language is visualwebarena written in?
- web-arena-x/visualwebarena is primarily written in Python.
- How popular is visualwebarena?
- web-arena-x/visualwebarena has 481 stars on GitHub.
- Where can I find visualwebarena?
- web-arena-x/visualwebarena is on GitHub at https://github.com/web-arena-x/visualwebarena.