Exa
Editor's pick #1AI-native neural search API built for LLM agents, not SERP scraping
What is Exa?
Exa (formerly Metaphor, rebranded 2024) is an AI-native search API that uses neural retrieval rather than SERP scraping. Its search index is optimised for semantic relevance rather than keyword ranking, with an Instant mode delivering results in under 180ms and a highlights feature that reduces LLM token usage by up to 90%. In independent FRAMES benchmark testing Exa achieved 54.4% accuracy versus competitors at 44.5% and 21.6%. Exa is SOC 2 Type II certified with a Zero Data Retention option. Notable clients include Cursor, AWS, Databricks, Groq, HubSpot, Cognition (Devin AI), and Monday.com (per company website; independently unverifiable).
Exa was founded in 2022 and is headquartered in San Francisco, CA, USA. The firm employs 11–50 people and works primarily with clients in AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce sectors. Its primary differentiator is: Neural search built for agents — not a SERP scraper — with 90% token reduction via highlights and 54.4% FRAMES accuracy.
Exa tech stack and services
| Service area | Details |
|---|---|
| AI agent research workflows requiring semantically relevant results rather than keyword-ranked SERP pages | Available for AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce clients |
| LLM context retrieval where minimising token usage matters — highlights cuts context by up to 90% | Available for AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce clients |
| Company and people data enrichment using structured database of 70M+ companies | Available for AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce clients |
| Coding assistant grounding and technical documentation lookup for developer tools | Available for AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce clients |
| Real-time web event monitoring via Monitors API with webhook delivery and deduplication | Available for AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce clients |
Exa use cases
Short answer: Exa is best suited for aI agent and LLM developers needing semantic neural search with <180ms latency and token-efficient structured output.
| Use case | Industries | Approach |
|---|---|---|
| AI agent research workflows requiring semantically relevant results rather than keyword-ranked SERP pages | AI / LLM Workflows, Research & Academia | REST API, Python SDK |
| LLM context retrieval where minimising token usage matters — highlights cuts context by up to 90% | AI / LLM Workflows, Research & Academia | REST API, Python SDK |
| Company and people data enrichment using structured database of 70M+ companies | AI / LLM Workflows, Research & Academia | REST API, Python SDK |
| Coding assistant grounding and technical documentation lookup for developer tools | AI / LLM Workflows, Research & Academia | REST API, Python SDK |
| Real-time web event monitoring via Monitors API with webhook delivery and deduplication | AI / LLM Workflows, Research & Academia | REST API, Python SDK |
Exa pricing
Short answer: Exa uses a pay-as-you-go ($7/1k searches); free tier (20k req/month) pricing approach. Minimum engagement starts at $0 (free tier; 20,000 requests/month).
| Engagement model | Typical range | Best for |
|---|---|---|
| Free tier | Variable; depends on team size | Large programmes or team augmentation |
| Pay-as-you-go | Variable; depends on team size | Large programmes or team augmentation |
| Enterprise contract | Variable; depends on team size | Large programmes or team augmentation |
Exa pros and cons
| Advantages | Things to consider |
|---|---|
| +20,000 free searches/month — most generous free tier of any neural search API in this category | -$7/1K searches is expensive for bulk SERP volume use cases — DataForSEO is ~11,600× cheaper per query |
| +Highlights feature reduces downstream LLM token usage by up to 90%, materially cutting inference costs | -Deep Search ($12/1K) and Deep Reasoning ($15/1K) costs accumulate quickly for multi-step research workflows |
| +<180ms latency with Exa Instant mode — suitable for user-facing real-time agent applications | -Optimised for semantic relevance, not exact keyword SERP matching — not a drop-in replacement for rank-tracking tools |
| +54.4% accuracy on FRAMES benchmark vs 44.5% and 21.6% for next competitors in independent testing | |
| +SOC 2 Type II certified with Zero Data Retention option for compliance-sensitive pipelines | |
| +Native MCP server and OpenAI-compatible SDK — integrates into existing agent frameworks without glue code |
Exa vs alternatives
How Exa compares to the other top Web Search API providers.
| Company | Best for | Key difference | Rating | Compare |
|---|---|---|---|---|
| NewsCatcher API | AI agents and LLM workflows requiring high-recall web... | Pay-per-validated-result pricing with coverage-first search that scans 50K+ pages per job — not just top-ranked SERP results | 4.8 | Full comparison |
| SerpAPI | Enterprise teams needing legally protected, structured SERP data... | U.S. Legal Shield (up to $2M coverage) plus 50+ engine support and three compliance certifications in a single API | 4.4 | Full comparison |
| Tavily | AI agent developers needing a drop-in web search... | Only web search API with PII protection and prompt injection blocking built into the retrieval layer — no custom security middleware required | 4.3 | Full comparison |
| Brave Search API | Privacy-conscious AI applications and products that need a... | Only fully independent web index in this category — 30B+ pages crawled directly, not a Google/Bing scraper, with no Big Tech dependency | 4.2 | Full comparison |
| Serper | Cost-sensitive applications needing fast Google SERP results with... | Lowest barrier to entry of any Google SERP API — 2,500 free queries with no card required, and 10+ search verticals in a single endpoint | 4.1 | Full comparison |
| DataForSEO | SEO agencies, analytics platforms, and bulk data pipelines... | $0.0006 per SERP (standard queue) — the lowest published per-query price reviewed, with 700-result depth and Baidu/Naver/Seznam support | 3.9 | Full comparison |
Exa FAQ
What is Exa?
Exa (formerly Metaphor, rebranded 2024) is an AI-native search API that uses neural retrieval rather than SERP scraping. Its search index is optimised for semantic relevance rather than keyword ranking, with an Instant mode delivering results in under 180ms and a highlights feature that reduces LLM token usage by up to 90%. In independent FRAMES benchmark testing Exa achieved 54.4% accuracy versus competitors at 44.5% and 21.6%. Exa is SOC 2 Type II certified with a Zero Data Retention option. Notable clients include Cursor, AWS, Databricks, Groq, HubSpot, Cognition (Devin AI), and Monday.com (per company website; independently unverifiable).
How much does Exa charge?
Exa uses pay-as-you-go ($7/1k searches); free tier (20k req/month) pricing. Minimum engagement starts at $0 (free tier; 20,000 requests/month). A discovery call is required to get project-specific quotes.
What tech stack does Exa use?
Exa works with REST API, Python SDK, TypeScript SDK, MCP server, JSON output. Primary industries served include AI / LLM Workflows, Research & Academia, Developer Tools, E-commerce.
Is Exa right for enterprise?
AI agent and LLM developers needing semantic neural search with <180ms latency and token-efficient structured output. 11–50 team size. Key consideration: $7/1K searches is expensive for bulk SERP volume use cases — DataForSEO is ~11,600× cheaper per query.
What are the best Exa alternatives?
The best alternatives to Exa depend on your use case. Top options are:
- NewsCatcher API: pay-per-validated-result pricing with coverage-first search that scans 50k+ pages per job — not just top-ranked serp results
- SerpAPI: u.s. legal shield (up to $2m coverage) plus 50+ engine support and three compliance certifications in a single api
- Tavily: only web search api with pii protection and prompt injection blocking built into the retrieval layer — no custom security middleware required
Compare Exa with other Web Search API providers
Last reviewed: June 2026. Verify all details directly with Exa before making a decision.