Research at the frontier of web search

How Search Quality Shapes RL Outcomes

Training search agents on Exa instead of a Google baseline yields higher pass@k, faster, at lower cost.

Exa Highlights

Dense, query-specific excerpts let agents ground on the web while reading dramatically fewer tokens.

Introducing Deep Max: State-of-the-Art Agentic Search

Our highest-quality search endpoint. SOTA on DSQA, FRAMES, and HLE-Search, and up to 20x faster than competitors.

Composing a Search Engine

How we built Canon, a search pipeline orchestrator for full control over our search engine as it scales in complexity.

WebCode: Search Evals for Coding Agents

Open-source web search evaluations for coding agents.

Introducing Exa Deep: An Agent for Every Search

Our best agentic search endpoint — faster, cheaper, with structured outputs and field-level grounding

Introducing Exa Instant

The fastest search engine in the world. Sub-200ms latency

Introducing Exa's Company Search Benchmarks

exa-d: Data Framework to Process the Web

How do you store and retrieve information from the web in a database?

Meet the Exacluster

Our 18-node GPU cluster training our next-gen search models

Serving BM25 with 50% memory reduction using a novel encoding scheme

How we cut our BM25 index footprint in half at billions-document scale without sacrificing performance.

How we built a web-scale vector database

It uses clustering, matryoshka embeddings, binary quantization, and SIMD operations. Written in rust of course 🦀

Scaling our highlights server

Serving real-time embeddings at scale is challenging. To launch Exa Highlights, we 4X’ed throughput by migrating from Python to Rust.