EvoBrowseComp: 800 Multilingual Search Agent Questions from Live-Web Traversal

EvoBrowseComp is a benchmark of 400 English and 400 Chinese contamination-free complex questions synthesized via live-web traversal. The dataset, created by Krystalan, is designed to evaluate search agents on evolving knowledge. It was last updated on June 11, 2026.

Use Cases

• Benchmarking search agent performance based on complex, contamination-free questions.
• Evaluating multilingual QA systems based on parallel English and Chinese question sets.
• Training retrieval-augmented generation (RAG) models based on questions synthesized from live-web traversal.

Strengths

• Contains 800 total questions (400 English, 400 Chinese).
• Questions are described as contamination-free and synthesized via live-web traversal.

Limitations

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

Provenance

Source

Krystalan on Hugging Face.

Collection Method

Questions synthesized via live-web traversal.

Freshness

Last updated 2026-06-11 18:20:48; freshness should be verified.