M3Eval: Multi-Modal Memory Evaluation Benchmark

M³Eval is a benchmark for evaluating multi-modal models through cognitively-grounded video tasks, released by researchers from Peking University and the University of Wisconsin-Madison. The dataset was released on June 4, 2026, by the PKU-VaLuE-Lab. It is designed to assess memory and reasoning capabilities in video-based contexts.

Use Cases

• Benchmarking multi-modal model memory based on video tasks
• Evaluating AI performance on cognitively-grounded reasoning tasks
• Training models for video-based question answering and recall
• Researching the intersection of cognitive science and machine learning

Strengths

• Released by a research lab affiliated with Peking University and the University of Wisconsin-Madison
• Focuses on cognitively-grounded evaluation, a specific and defined research goal
• Released on June 4, 2026, indicating recent development

Limitations

• Column-level documentation and sample data are unavailable, making field semantics unclear
• The dataset size, row count, and file formats are unknown, limiting suitability assessment
• The license is unspecified, which may restrict usage

Provenance

Source

PKU-VaLuE-Lab (Peking University)

Freshness

Last updated 2026-06-04 02:52:00