Embench<br>Which retrieval stack actually works on your data, for your agent?<br>Embench is a retrieval lab. Index a corpus, run semantic, keyword, hybrid, and reranked search side by side, and see precision, recall, and MRR inline — before your agent depends on it.<br>Try the labFree to use, no signup — quota per browser session.<br>Explore in the browser, then hand the same harness to your agent: every run here is a plain REST call with a stable JSON contract. CLI for local and CI evals is on the roadmap.<br>Side-by-side retrieval<br>Semantic, BM25 keyword, grep, hybrid, and reranked search on the same corpus, in one run.

Evals built in<br>Mark expected docs and get precision, recall, and MRR next to every result list.

Open-source models<br>MiniLM, BGE, Qwen, and Stella embeddings plus cross-encoder rerankers. No setup.

Join the waitlistHosted API, saved corpora, batch jobs, and a CLI are on the way.

Loading available models…<br>Embed & SearchEmbedding ComparisonReranker Comparison<br>Documents (one per line, or `doc-id | text`)3 docs to index<br>Vector databases store embeddings for fast semantic search.<br>Rerankers rescore retrieved documents using the query and document text.<br>Embeddings map text into numeric vectors that capture meaning.<br>Split / ChunkNewlineBlank lineFixed size

Chunk Sizeinfo

Overlapinfo

Embedding Models1 selectedsvg]:size-3 [&>svg]:pointer-events-none focus-visible:border-ring focus-visible:ring-ring/50 focus-visible:ring-[3px] aria-invalid:ring-destructive/20 dark:aria-invalid:ring-destructive/40 aria-invalid:border-destructive transition-[color,box-shadow] overflow-hidden bg-secondary text-secondary-foreground [a&]:hover:bg-secondary/90 gap-1 pr-1">all-MiniLM-L6-v2

Index Corpus

Query<br>Top K<br>Search<br>Retrieval ModesSemanticKeyword (BM25)GrepHybridSemantic + Reranker

Metricinfo<br>Cosine Similarity<br>Hybrid Weight0.50<br>KeywordSemantic

Reranker ModelsSelect models...

Expected Relevant Doc IDsUsed for inline metrics. You can also mark results after searching.