Exa AI ships a neural search API built for LLM applications, AI agents, and RAG pipelines. Queries retrieve pages by meaning rather than keyword density. The Websets product layers B2B lead research on top of the same search infrastructure.
No consumer interface. Developer-only product.
Recent developments
- May 12, 2026: Anthropic launched Claude for Legal with Microsoft 365 plug-ins and 20-plus practice-area MCP connectors. For Exa, the read is that enterprise retrieval is increasingly bundled into the chat assistant via MCP, not procured separately. Exa benefits when teams want a generic neural retrieval primitive across providers, and loses to bundled retrieval when the workflow lives entirely inside one assistant.
- March 3, 2026: Exa simplified search pricing: contents for the first 10 results are now included with
searchat no extra charge, Deep Search dropped 20% to $12 per 1,000, and a new Deep Search (Reasoning) tier landed at $15 per 1,000. Most agent stacks no longer pay a separate contents charge for primary searches.
System Verdict
Pick Exa if the app needs semantic web retrieval inside an LLM pipeline. The neural index returns conceptually relevant pages when terminology differs between query and source. LangChain and LlamaIndex ship native connectors, so integration is measured in minutes, not days.
Skip it for consumer search, keyword-only lookups, or jobs that need Google-scale coverage. Obscure, non-English, and very recent pages are patchier than SerpAPI. Index breadth is the trade-off for semantic ranking.
Who pays which tier: Free tier (1,000 requests/month) for prototyping, pay-as-you-go core search for most production agents, Websets Starter ($49/mo) for B2B research teams, Websets Pro ($449/mo) for scaled lead-gen and enterprise workflows.
Key Facts
| Product type | Neural search API for developers |
| Search | $7 per 1,000 requests (10 results, contents + highlights bundled) |
| Deep Search | $12 per 1,000 requests (multi-step research) |
| Deep Search (Reasoning) | $15 per 1,000 requests |
| Monitors | $15 per 1,000 requests (long-running watch jobs) |
| Answer | $5 per 1,000 requests (cited answer endpoint) |
| AI Page Summaries | $1 per 1,000 pages across all endpoints |
| Additional results | $1 per 1,000 results beyond 10 |
| Free tier | 1,000 requests/month, full API access |
| Websets Starter | $49/mo: 8,000 credits, 1 seat, 100 results/Webset |
| Websets Pro | $449/mo: 100,000 credits, 10 seats, 1,000 results/Webset |
| Enterprise | Custom volume pricing, SLA, security |
| Framework integrations | LangChain, LlamaIndex (native connectors), MCP-compatible |
| Funding | Series B, $85M (2025) |
Every data point above was verified against vendor documentation on 2026-05-13. See Sources.
What it actually is
One developer product with two surfaces. The core API now ships five paid endpoints (search, deep_search, answer, monitors, contents) plus an AI Page Summaries add-on. Search returns neural results with full page text and highlights bundled for the first 10 hits. Deep Search runs multi-step research with optional reasoning. Monitors watches a query and fires when new matches appear. Answer ships cited responses for question-style retrieval. Websets sits on top and uses the same index for B2B research, verifying every row against user-specified criteria.
The March 3, 2026 pricing update bundled content extraction into base search cost. Most agents no longer pay a separate scrape fee.
The moats: a neural index trained for LLM-style queries, native framework integrations, MCP compatibility, and Websets as a second revenue surface that locks research teams in. The weakness is index breadth. Exa is narrower than Google on obscure domains and very recent pages.
When to pick Exa
- Building an AI agent that needs live web context. Semantic ranking returns intent-matched pages without heavy post-filtering.
- Wiring RAG to dynamic web sources. Contents endpoint delivers clean page text up to 20k tokens. No separate scraper layer needed.
- Running B2B lead research or market mapping. Websets verifies every result against criteria, reducing false positives that kill manual list-building.
- Shipping inside a LangChain or LlamaIndex stack. Drop-in connectors mean no glue code.
- Cost-sensitive prototyping. $7 per 1,000 requests undercuts SerpAPI at agent scale.
When to pick something else
- Consumer search UI for end users: Perplexity or Kagi. Exa has no front end.
- Maximum index breadth on obscure or non-English pages: SerpAPI (Google-backed).
- Paid-search quality without building an app: Kagi.
- Academic paper retrieval specifically: Consensus or Elicit.
- Cited answer engine for research tasks: Perplexity or You.com.
Pricing
Pricing via exa.ai/pricing:
| Plan / Endpoint | Price | What you get |
|---|---|---|
| Free | $0 | 1,000 requests/month, full API access |
| Search | $7 / 1K requests | 10 neural results with contents + highlights bundled; $1/1K for extra results |
| Deep Search | $12 / 1K requests | Multi-step neural research |
| Deep Search (Reasoning) | $15 / 1K requests | Adds reasoning over multi-step results |
| Answer | $5 / 1K requests | Cited answer endpoint |
| Monitors | $15 / 1K requests | Long-running watch jobs |
| Contents endpoint | $1 / 1K pages | Standalone page retrieval |
| AI Page Summaries | $1 / 1K pages | Cross-endpoint summary add-on |
| Websets Starter | $49/mo | 8,000 credits, 1 seat, 100 results/Webset, 2 concurrent |
| Websets Pro | $449/mo | 100,000 credits, 10 seats, 1,000 results/Webset, 10 concurrent |
| Enterprise | Custom | Volume pricing, SLA, security review |
Prices verified 2026-05-13 via Exa pricing and Exa pricing update. Startups and education projects can apply for $1,000 in free credits.
Against the alternatives
| Exa AI | SerpAPI | Tavily | |
|---|---|---|---|
| Ranking approach | Neural semantic | Google keyword | Agent-optimized mix |
| Index breadth | Narrower | Google-scale | Mid |
| Content retrieval | Bundled (contents endpoint) | Separate scrape | Bundled |
| Free tier | 1,000 requests/month | Limited trial | Monthly free quota |
| Framework integrations | LangChain, LlamaIndex native | Community wrappers | LangChain native |
| B2B research product | Websets | None | None |
| Best viewed as | Semantic retrieval for LLMs | Google proxy | Agent-tuned search |
Failure modes
- Index breadth is narrower than Google. Obscure domains, non-English content, and very recent pages can be missing. Production pipelines sometimes pair Exa with SerpAPI as a fallback.
- No consumer UI. End users need an application built on top. Not a Google replacement.
- Latency spikes under load. Neural indexing is compute-heavy. Average 800ms is competitive, but peaks happen.
- Pay-as-you-go cost can grow unnoticed. High-traffic apps need usage alerts. Volume pricing is worth negotiating once past 100K requests/month.
- Thinner ecosystem than SerpAPI. Fewer tutorials, fewer third-party wrappers, smaller Stack Overflow footprint.
- Websets credits expire. Unused Starter credits do not roll over indefinitely. Teams with bursty usage should size Pro against peak month, not average.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility x Value x Moat x Longevity, unweighted average). Last verified 2026-05-13 against Exa pricing, Exa pricing update, and Exa Series B announcement.
FAQ
Is Exa free? Yes, for prototyping. The free tier includes 1,000 requests per month with full API access. Pay-as-you-go Search starts at $7 per 1,000 requests, Deep Search at $12, Answer at $5, and Monitors at $15. AI Page Summaries are $1 per 1,000 pages across endpoints.
What is Websets? A B2B research product built on the Exa index. Users describe criteria in natural language. AI agents verify each result before returning it. Starter is $49/mo; Pro is $449/mo.
Exa vs SerpAPI? SerpAPI proxies Google keyword results. Exa runs a neural index trained on LLM-style queries. Pick SerpAPI for breadth and exact Google parity. Pick Exa for semantic intent matching and bundled content retrieval.
Does Exa work with LangChain and LlamaIndex? Yes. Both frameworks ship native Exa connectors. Drop-in tool for agent and RAG workflows.
What happened to Metaphor Systems? Exa is the rebrand of Metaphor Systems from 2023. Same neural search API, broader positioning toward the AI developer market.
Sources
- Exa pricing: current core API and Websets pricing
- Exa pricing update: March 2026 pricing simplification notes
- Exa Series B announcement: $85M raise, product direction
- Exa API documentation: full API reference and integration guides
Related
- Category: AI Search
- Comparisons: Exa AI vs Kagi | Exa AI vs Perplexity | Exa AI vs Phind | Exa AI vs You.com
