Common-O: Multi-Scene Reasoning with 10,000+ Household Objects

Common-O contains between 10,000 and 100,000 image-text pairs designed by Meta researchers in 2026 to evaluate multimodal LLM reasoning. The data is organized into two subsets featuring household objects to test the ability of models to identify common elements across 3 to 16 different scenes.

Use Cases

• Evaluating multimodal LLMs on multi-scene reasoning tasks using the 'what's in common' query feature
• Benchmarking model performance degradation when moving from the standard subset to the 'Common-O Complex' subset
• Fine-tuning vision-language models for relational reasoning across multiple household object images

Strengths

• Includes a 'Common-O Complex' subset with 8-16 objects to test reasoning scalability
• MIT licensed for open research use
• Large scale with 10,000 to 100,000 reasoning pairs

Limitations

• Domain-specific focus on household objects may limit generalizability to non-domestic environments
• Lack of explicit column documentation in metadata

Provenance

Source

Facebook (Meta) AI Research, associated with Arxiv paper 2511.03768

Freshness

Last updated March 2026.