AgentVista: Evaluating Multimodal Agents inUltra-Challenging Realistic Visual Scenarios Real-world multimodal agents solve multi-step workflows grounded in visual evidence. AgentVista is the first benchmark that evaluates multimodal agents on realistic, ultra-challenging visual scenarios requiring long-horizon hybrid tool use. It spans 25 sub-domains across 7 categories with 209 expert-curated tasks. Even the best model, Gemini-3-Pro, achieves only 27.3% overall accuracy. This repository provides a lightweight yet general agent framework for reproducible evaluation. <i
| Stars | 34 |
| Forks | 3 |
| Language | Python |
| Category | AI Tool |
| Quality Score | 65.8238092922779/100 |
| Last Updated | 2026-03-10 |
| Created | 2026-02-26 |
| Platforms | python |
| Est. Tokens | ~200k |
Explore other popular ai tool tools:
AgentVista is Benchmarking multimodal agents on realistic, ultra-challenging visual scenarios requiring long-horizon hybrid tool use.. It is categorized as a AI Tool with 34 GitHub stars.
AgentVista is primarily written in Python.
You can find installation instructions and usage details in the AgentVista GitHub repository at github.com/hkust-nlp/AgentVista. The project has 34 stars and 3 forks, indicating an active community.