When would you reach for a cross-encoder over a bi-encoder?

Attempt it

Your answer

Quick reference

TL;DR

A cross-encoder scores (query, doc) jointly in one transformer pass; you use it to rerank a small candidate set, never to search the full corpus.

Memory aid

Sign in for the mnemonic that makes this stick. See Pro pricing.

Easy to grasp

Think of dating apps. The bi-encoder is like writing a one-page profile for every person on the planet, then sorting profiles by how well yours matches at a glance. That is fast but shallow. The cross-encoder is the actual first date: you sit across from one person and pay full attention to the back and forth. Way more accurate, but you only have time for a handful of dates. So you screen with profiles first (bi-encoder retrieves top 50), then go on dates with just the finalists (cross-encoder reranks). The cross-encoder never meets the whole planet.

Key concepts

Embeddings

Detailed answer & concept explanation

~5 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.

Sign in for 5 free unlocks/day or go Pro for unlimited. See Pro pricing.

Quick scan · 2–4 min

Everything important, quickly.

6 to 8 min: cross-encoder definition + why O(N) scoring blocks first-stage use + two-stage pattern + latency math + decision rule + 2026 reranker landscape.

Real-world examples

Real products, models, and research that use this idea.

Cohere rerank-3.5 (2025) is the managed-API default for production reranking in 2026.
Voyage rerank-2.5 pairs naturally with voyage-3-large embeddings in Anthropic-aligned stacks.
BGE-reranker-v2-m3 from BAAI is the most-deployed open-weight cross-encoder for multilingual rerank.
Anthropic's Claude documentation explicitly recommends a two-stage retrieve + rerank pattern for production RAG.
ColBERT v2 from Stanford remains the canonical late-interaction middle ground between bi- and cross-encoders.

Interviewer follow-ups

What an interviewer would ask next. Try answering before peeking at the approach.

QHow would you choose between Cohere rerank-3.5, Voyage rerank-2.5, and BGE-reranker-v2-m3 for a new project?

Sign in or go Pro to see the approach for this follow-up.

QWhy is ColBERT often described as a middle ground between bi- and cross-encoders?

Sign in or go Pro to see the approach for this follow-up.

QWhat does recall@100 being your reranker ceiling actually mean in practice?

Sign in or go Pro to see the approach for this follow-up.

Don't say this

Red flags and common mistakes that signal junior thinking. Click to expand.

Most common mistake

Saying you could use a cross-encoder as a first-stage retriever if you precompute scores. You cannot, the query is new at search time, so every pair must be scored fresh.

Sign in or go Pro to see all red flags and common mistakes. See Pro pricing.

60-second cheat sheet

The night-before-the-interview bullets. Scan these on the way to the call.

Sign in for 5 unlocks/day or go Pro for unlimited revision sheets. See Pro pricing.

References

Primary sources. Skim if you want the original framing.