An AI-native search infrastructure company that builds its own search engine from scratch—crawler, index, embedding models, and vector database—rather than wrapping other providers. It tracks 500B+ URLs (some sources cite 2T), serves 1B+ queries monthly, and powers search for Cursor, Cognition, HubSpot, OpenRouter, Monday.com and 500K+ developers . Its Exa 3.0 release (October 2026) ranks #1 on the Artificial Analysis Search Index for quality, requiring 20–50% fewer search queries per task than competitors . Core products: Search (Lite $4/1k, Auto $7/1k), Deep Search ($12–15/1k), **Contents** ($1/1k pages), Answer ($5/1k), **Monitors** ($15/1k), and Agent API ($0.012–$1.00/run) . Pricing is pay-as-you-go with no subscription or minimum; the Free tier gives $10 credits/month** (resets monthly, no card required), plus a **keyless hosted MCP server** for evaluation . Raised **$250M Series C at a $2.2B valuation (May 2026, a16z led) .

Pros & Cons:

Pros: The self-built search stack is the core differentiator—Exa owns its crawler, index, and embedding models, which it claims enables higher quality, lower latency (sub-200ms), and token efficiency (95% fewer tokens via Dynamic Highlights) than providers that wrap other engines ; the #1 ranking on Artificial Analysis Search Index (independent benchmark) and 20–50% fewer queries per task provide external validation of quality claims ; pricing is transparent and low-friction—pay-as-you-go from $4/1k searches, $10 monthly free credits with no card required, and a keyless MCP server for immediate testing ; the customer list (Cursor, Cognition, HubSpot) and 500K+ developer adoption are strong production signals.

Cons: The review trail on major software review sites is thin—only 1 G2 review surfaced, so independent evidence beyond Exa’s own benchmarks and named customers is limited ; usage-based pricing across multiple meters (search tier, extra results, contents, agent effort levels, third-party data via Exa Connect) makes monthly cost depend heavily on how an agent calls the API, requiring careful monitoring ; it is developer infrastructure only—there is no end-user research app, and Zero Data Retention, HIPAA, and SSO are reserved for Enterprise plans ; the /findSimilar endpoint is deprecated and /research was retired in April 2026 in favor of deep-reasoning search type, so teams migrating from older integrations need to update their code ; the company’s rapid growth and $2.2B valuation mean expectations are high, but it remains a young company relative to established search infrastructure providers.

Who is it for?:Developers building AI agents, copilots, or RAG pipelines that need high-quality, low-latency web search via API; teams already using Claude Code, Cursor, or other MCP-enabled tools that want live web search without managing API keys (via the keyless MCP server); organizations that need deep research, list-building, or enrichment workflows (Agent API, Monitors); enterprises requiring custom indexes, volume discounts, SLAs, or Zero Data Retention. Not suited for non-developers seeking a consumer search/chat product (Exa has no end-user app), teams that need HIPAA or SSO on a non-Enterprise plan, or those wanting a fully open-source or self-hosted search stack.