Knowledge-Grounded Image Generation Benchmark for Agentic Search

KnowGen Bench provides evaluation data for Gen-Searcher, a multimodal agent for image generation requiring complex real-world knowledge. The benchmark supports training and testing agents that search the web, browse evidence, and reason over multiple sources. It was created by the GenSearcher team and last updated in April 2026.

Use Cases

• Benchmarking agent performance on knowledge-grounded image generation tasks.
• Training multimodal agents to search for and reason over web evidence before generation.
• Evaluating the ability of models to integrate complex real-world knowledge from multiple sources.

Strengths

• Designed for the first multimodal deep research agent for knowledge-intensive image generation.
• Benchmark data is associated with a published paper (arXiv:2603.28767).

Limitations

• Specific dataset size, row count, and column details are unknown.
• The scope and diversity of the included knowledge domains are not specified.

Provenance

Source

GenSearcher team via Hugging Face.

Collection Method

Created as a benchmark for the Gen-Searcher research project.

Freshness

Last updated in April 2026.