Pointerbench: A GUI Grounding Benchmark Suite for Computer-Use Models

Pointerbench is a small GUI grounding benchmark suite for computer-use models created by WarmwindOS. Each example includes a screenshot, an instruction, target geometry in absolute pixels, and a binary evaluation rule. The dataset was last updated on July 2, 2026.

Use Cases

• Benchmarking model performance on GUI grounding tasks based on screenshot-instruction pairs.
• Training models to interpret on-screen instructions and locate target geometry based on pixel coordinates.
• Evaluating the precision of computer-use models using binary evaluation rules provided for each example.

Strengths

• Provides a structured benchmark with three distinct subsets, each testing different GUI interaction capabilities.
• Includes binary evaluation rules for each example, enabling clear performance measurement.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

WarmwindOS

Freshness

Last updated 2026-07-02 18:55:40; freshness should be verified.