Exa

Market position

Exa positions itself as the highest-quality search API at every latency point and as one of the few independent, non-wrapper search engines, contrasting itself with providers that resell traditional engines. Vendor-published benchmarks on the homepage show Exa ahead of Perplexity and Brave on FRAMES (54.4% vs 44.5% and 21.6%), Tip-of-Tongue (54.2% vs 36.7% and 19.3%) and Seal0 (36% vs 34.5% and 8.2%). Exa Agent is benchmarked against Perplexity Agent tiers, Parallel Task tiers and frontier models on WideSearch, BrowseComp, DeepSearchQA, FinanceAgent-V2, People-FindAll and Company-FindAll. These figures are self-reported and were not independently verified in the sources reviewed.

Independently built index and retrieval stack rather than a wrapper over other search engines; ranking optimized for information quality with no advertising; full page content and token-efficient highlights rather than URLs and titles alone; configurable latency-quality tiers from sub-second to multi-minute deep research; vertical indexes over people, companies, publications, news and code; customization such as excluding thousands of domains, high result counts and custom classifiers; and zero data retention plus SOC 2 Type II, GDPR, CCPA and HIPAA-oriented controls.

Technology

Exa built its search stack from scratch rather than wrapping third-party engines, which it says enables true zero data retention and control over quality, latency and cost. Components include large-scale crawlers tracking 500B+ URLs (1.4 trillion URLs tracked per the homepage) and crawling billions of documents per day, purpose-built vector databases serving 100 billion documents at 10k+ QPS, and in-house embedding and agentic models trained on clusters of hundreds of H200s and a B200 training cluster. Specialized models include finetuned code-search embeddings and a highlights model that extracts only relevant excerpts from pages. The pipeline described on the About page includes query enrichment, tokenization, entity extraction, web/inverted/news indexes, reranking, snippet extraction and safety filtering. Exa states it is on track to exceed Google-scale traffic and index size in 2027.

Go-to-market

Developer-led self-serve adoption through free API trials, an API playground, SDKs (exa-py, exa-js), dashboard onboarding and documentation aimed at both humans and coding agents, combined with an enterprise sales motion. Following the Series C, Exa hired Marcus Holm, President of LaunchDarkly, as CRO to build a global GTM organization and is hiring across sales, solutions and marketing.

AI application developers and agent builders — coding agents, chat and voice applications, deep-research tools, go-to-market and finance agents — plus enterprises with compliance requirements. Named users include Cursor, Cognition (Devin), HubSpot, OpenRouter, Monday.com, Notion AI, and private equity, consulting and financial firms. Firefox is cited as a search distribution partner.

Geography

Headquartered in San Francisco; the company describes hiring worldwide and building a global go-to-market organization, and supports EU (GDPR) and California (CCPA) data-protection requirements.

History

Founded 2021; first search engine launched November 2022; first web search API for AI announced in early 2023. In September 2025 Exa raised an $85M Series B led by Benchmark at a $700M valuation, with Lightspeed, Y Combinator and NVIDIA's NVentures participating and Peter Fenton joining the board. In May 2026 it raised a $250M Series C at a $2.2B valuation led by a16z, alongside senior hires including a CRO from LaunchDarkly and engineers from Meta, Yandex and Google. Product and distribution milestones in 2026 include Exa Agent (June), state-of-the-art academic publication search (July) and powering search in Firefox (August).

Risks & controversies

Most performance, index-size and customer-count claims come from Exa's own website and blog and are not independently verified in the material reviewed; figures also vary between pages (for example 500,000 vs 400,000+ developers, 70M+ vs 50M+ companies, 1.4T vs 500B+ URLs, and sub-180ms vs ~250ms instant latency). The business depends on continued large-scale crawling of the public web and on capital-intensive GPU and indexing infrastructure, and the company competes with far larger incumbent search providers as well as AI search vendors.