MM-ToolSandBox: A Unified Framework for Evaluating Visual Tool-Calling Agents

Source

Source name: arXiv
Source kind: paper
Authors / organization: Kaixin Ma, Di Feng, Alexander Metz, Jiarui Lu, Eshan Verma, Afshin Dehghan
Venue: arXiv: cs.CV
Published: 2026-07-13
Collected: 2026-07-14
URL: http://arxiv.org/abs/2607.11818v1
PDF: https://arxiv.org/pdf/2607.11818v1
DOI: UNKNOWN

Plain-Language Brief

The paper introduces MM-ToolSandBox, a new evaluation framework for visual tool-calling agents. It supports multi-image, multi-turn tasks and offers a stateful execution environment with over 500 tools across 16 application domains. The framework produces diverse scenarios through automated generation and human verification. Evaluating various models on this benchmark reveals that current models lack robust visual tool-calling capability, with even the best model achieving below 50% success rate. The main challenge is found to be inaccurate information extraction from images, not only planning.

What the Document Says

The MM-ToolSandBox framework is presented for evaluating visually grounded tool-calling agents.
It provides a stateful execution environment with over 500 tools across 16 application domains.
The framework supports multi-image, multi-turn tasks and handles realistic conversational phenomena like goal revisions, error corrections, and state mutations.
An automated scenario generation pipeline produces diverse scenarios through information-flow-guided planning and quality filtering.
The pipeline yields 258 human-verified nominal scenarios and 50 variants targeting interactive UI applications.
Evaluation of 12 state-of-the-art models shows that current models lack robust visual tool-calling capability.
Even the best model achieves below 50% success rate.

Key Concepts Mentioned

MM-ToolSandBox
evaluation framework
visually grounded tool-calling agents
stateful execution environment
multi-image, multi-turn tasks
information-flow-guided planning
realistic conversational phenomena
goal revisions
error corrections
state mutations
scenario generation pipeline
human verification
interactive UI applications
current models
robust visual tool-calling capability
success rate

Claims Made by the Document

MM-ToolSandBox offers a new benchmark and evaluation framework for visually grounded tool-calling agents.
Current models lack robust visual tool-calling capability.

Evidence Used

Introduction of MM-ToolSandBox
Results of model evaluations

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

We introduce MM-ToolSandBox, a benchmark and evaluation framework for visually grounded tool-calling agents. The framework provides a stateful execution environment spanning 500+ tools across 16 application domains, supporting multi-image, multi-turn tasks where agents must ground progressively arriving visual inputs into executable tool calls while handling realistic conversational phenomena (goal revisions, error corrections, state mutations). An automated scenario generation pipeline produces diverse, visually grounded scenarios through information-flow-guided planning and multi-stage quality filtering, yielding 258 human-verified nominal scenarios and 50 variants targeting interactive UI applications. Evaluating 12 state-of-the-art models, from 4B open-weight to frontier proprietary systems, shows that current models still lack robust visual tool-calling capability: even the best model achieves below 50% success rate. Our failure analysis further reveals that visual precision, not only planning, is a primary bottleneck for capable models: 53% of failures stem from incorrect information extraction from images despite otherwise correct task workflows. A planning-to-precision crossover emerges with scale: smaller models fail at deciding what to do, while larger models fail at perceiving what they see, suggesting fundamentally different research directions for improving models at different capability levels. The framework and the benchmark are publicly available at https://github.com/apple/ml-mmtoolsandbox