PPT-Eval: A Benchmark for Computer-Use Agents on PowerPoint Tasks

2026-06-30

AI Summary

Existing agents lack fine-grained evaluation methodologies for complex, multimodal PowerPoint tasks, making it difficult to assess partial completion and diverse correct solutions. This work proposes the first fine-grained evaluation framework specifically designed for PowerPoint manipulation, introducing a benchmark comprising 120 tasks and a multidimensional scoring mechanism based on human-designed rubrics. The framework supports partial credit for content creation and editing tasks, incorporates aesthetic penalties, detects redundant operations, and provides natural language feedback. It achieves a Kendall’s τ-b correlation of 0.77 with human judgments. Experimental results reveal that even state-of-the-art models, such as Claude-4.5-Opus, attain only a 45% full-task success rate and a 57% average partial score, highlighting significant limitations in current agent capabilities.

Abstract

Creating and editing slides is a rich, multimodal activity that is ubiquitous in professional and educational settings, making it an ideal testbed for real-world computer-use agents. Microsoft PowerPoint is among the most widely adopted and feature-rich environments for presentation creation. We introduce PPT-Eval, a benchmark of 120 PowerPoint tasks across 12 files that cover both content creation and presentation editing scenarios, organized by difficulty. A central challenge in this domain is evaluation: tasks are complex, multimodal, and often admit many valid solutions. Moreover, today's agents frequently make only partial progress, which binary success metrics fail to capture. To address this, we design a robust evaluation framework to help create task-specific rubrics for PowerPoint tasks, taking inspiration from and building on past works for rubric-based evaluation. These rubrics award partial credit for intermediate steps, penalize unnecessary changes and poor aesthetics, and provide natural language feedback. This nuanced approach proves highly effective, achieving a Kendall's τ-b correlation of 0.77 with human judgments. We find that existing frontier agents still struggle with solving PowerPoint tasks, with strong models like Claude-4.5-Opus achieving only a 45% success rate and an average partial score of 57%. The benchmark is located at: https://microsoft.github.io/ppteval.

Problem

• computer-use agents
• PowerPoint tasks
• multimodal evaluation
• partial credit
• rubric-based assessment

Innovation

• rubric-based evaluation
• multimodal benchmark
• partial credit scoring
• computer-use agents
• PowerPoint automation