# Exa — The world's data for AI agents

> Search the largest index of public web and private information.

Introducing Exa 3.0: Our largest ever improvement to search quality. [Learn more](https://exa.ai/blog/exa-3).

For the complete site-wide index, see [llms.txt](https://exa.ai/llms.txt).

Join 500,000 developers building AI on top of the world's information.

- [Try the API for free](https://dashboard.exa.ai)
- [API docs](https://exa.ai/docs)
- [API Playground](https://dashboard.exa.ai/playground/search)
- [Pricing](https://exa.ai/pricing)
- [Contact sales](https://exa.ai/contact/sales)

## Products

- [Search API](https://exa.ai/products/search)
- [Contents API](https://exa.ai/docs/contents/quickstart)
- [Agent API](https://exa.ai/products/agent)
- [Monitors API](https://exa.ai/docs/monitors/quickstart)
- [Websets](https://exa.ai/websets)
- [Exa Connect](https://exa.ai/connect)
- [MCP Server](https://exa.ai/mcp)
- [Integrations](https://exa.ai/integrations)

## Exa leads across benchmarks

Exa 3.0 tops public evals and the verticals customers rely on most

### Agentic search

Exa 3.0 leads on BrowseComp, DSQA, and WideSearch, the standard benchmarks for search agents.

| Benchmark (accuracy %) | Exa 3.0 | Perplexity | Brave | Parallel |
| --- | ---: | ---: | ---: | ---: |
| [BrowseComp](https://openai.com/index/browsecomp/) | 91.8 | 91.5 | 83.5 | 72.7 |
| DSQA | 88.2 | 87.6 | 78.6 | 85.3 |
| [WideSearch](https://arxiv.org/abs/2508.07999) | 56.1 | 54.0 | 54.8 | 55.4 |

### Search verticals

Best-in-class across code, finance, and GTM intelligence, not just general web queries.

| Benchmark (accuracy %) | Exa 3.0 | Perplexity | Brave | Parallel |
| --- | ---: | ---: | ---: | ---: |
| [WebCode](https://exa.ai/blog/webcode) | 85.5 | 73.9 | 69.4 | 76.3 |
| FinSearchComp | 71.4 | 63.4 | 57.6 | 68.0 |
| [ATLAS](https://exa.ai/blog/atlas-benchmark) | 44.5 | 37.3 | 27.7 | 32.5 |

### FindAll

Finds more of the companies and people that match a list-building query.

| Benchmark (score) | Exa 3.0 | Perplexity | Brave | Parallel |
| --- | ---: | ---: | ---: | ---: |
| Company FindAll | 15.4 | 14.2 | 13.3 | 9.6 |
| People FindAll | 7.6 | 6.8 | 3.8 | 3.6 |

### Cost

Exa 3.0 Lite and Auto set the new Pareto frontier for search.

| Provider | Agentic search score | Cost per task |
| --- | ---: | ---: |
| Exa 3.0 Auto | 80.3 | $0.0656 |
| Exa 3.0 Lite | 79.2 | $0.0444 |
| Perplexity Medium | 80.0 | $0.0703 |
| Perplexity High | 79.0 | $0.0666 |
| Parallel Advanced | 74.8 | $0.0592 |
| Parallel Basic | 73.2 | $0.0668 |
| Brave LLM Context | 74.6 | $0.0794 |

[Full evaluations](https://exa.ai/evals)

## Designed to make your agent smarter

- Web-scale index: comprehensive coverage over all [search verticals](https://exa.ai/products/search) so your agent has the right context to solve its problem.
- The docs your coding agent needs: powers coding agents with smart, token-efficient search across docs and repos.
- Token-efficient contents with highlights: a specialized model trained to [extract the most relevant excerpts](https://exa.ai/docs/search/highlights) from the web. Up to 25T tokens passed to models each week are concise excerpts powered by highlights.
- Low-latency to deep research in one API: configure for fast results or [deep research](https://exa.ai/products/deep) with structured outputs and grounded citations

## Enterprise

[Enterprise-grade security and controls](https://exa.ai/enterprise):

- SOC 2 Type II — independently audited controls for security, availability, and confidentiality.
- GDPR — EU data-subject rights honored, with lawful-basis handling for personal data.
- CCPA — California consumer privacy rights, including access and deletion requests.
- HIPAA — safeguards for protected health information, with a BAA available.
- Zero data retention — your queries and results are never stored or trained on.
- Enterprise agreements — BAA discussions, DPA documentation, and custom data-security arrangements.

[Security](https://exa.ai/security) · [Trust Center](https://trust.exa.ai)

## Training retrieval models at massive scale

We train embedding and agentic models for state-of-the-art retrieval, crawl billions of documents per day, and build vector databases from scratch for 10k+ QPS. See [Research](https://exa.ai/research).

## Trusted by world-class teams

[Customer stories](https://exa.ai/customers) — including Cognition, StackAI, Firefox, HubSpot, 11x, CodeRabbit, and monday.com.

## More

- [Blog](https://exa.ai/blog)
- [Careers](https://exa.ai/careers)
- [About](https://exa.ai/about)
- [API status](https://status.exa.ai)
- [FAQ](https://exa.ai/docs/admin/faqs)
- [Contact](https://exa.ai/contact)
- [Site index for agents](https://exa.ai/llms.txt)