--- source_url: "https://exa.ai/versus/tavily" title: "Exa vs Tavily: Search API Benchmarks and Comparison" mirrored_at: 2026-09-02T01:02:49.146Z host: exa.ai cited_in_42a: true mirror_canonical: "https://index.42a.ai/exa.ai/versus/tavily" --- > **Original source:** https://exa.ai/versus/tavily Last updated August 4, 2026 A comprehensive evaluation of Exa versus Tavily, focusing on specialized entity retrieval, response times, and optimized token usage for agentic workflows. Exa and Tavily are search API providers designed for agentic workflows rather than human browsing. Both systems deliver ranked results with token-succinct content, offer Python and JavaScript SDKs, and support MCP (Model Context Protocol) servers. Exa is the preferred choice for retrieval that is not a web page: people, companies, code, and academic publications, plus agent loops where tail latency sets the timeout budget. The Tavily API is a better fit when you want full page text on every result, bundled into the search price, and for high-volume general web lookups. This comparison is grounded in empirical benchmarks. Each metric specified below includes its source dataset, query volume, and evaluation date. Tavily was acquired by Nebius, an Amsterdam-based AI cloud company, in February 2026, and is being integrated into that company's cloud platform. Exa engineered its search infrastructure and high-precision retrieval models completely from scratch. As of August 2026, Exa's CEO [reported](https://x.com/WilliamBryk/status/2084321018546704624) that their index tracks 1.4T URLs and serves 100B pages, establishing it as one of the largest operational indexes in the industry. That difference shows up in the results below. Exa exposes people, companies, code, and publications as categories on its own index rather than leaving them to general web retrieval, maintaining full control over its product roadmap. On this page - [Exa vs Tavily search API benchmarks](#exa-vs-tavily-search-api-benchmarks) - [Best use cases for the Exa search API](#best-use-cases-for-the-exa-search-api) - [Best use cases for the Tavily API](#best-use-cases-for-the-tavily-api) - [Exa search API platform overview](#exa-search-api-platform-overview) - [How to choose between the Exa and Tavily APIs](#how-to-choose-between-the-exa-and-tavily-apis) - [Switching from the Tavily API to Exa](#switching-from-the-tavily-api-to-exa) - [Exa vs Tavily FAQ](#exa-vs-tavily-faq) ## Exa vs Tavily search API benchmarks The datasets, grading logic, and provider adapters are all open source at [github.com/exa-labs/benchmarks](https://github.com/exa-labs/benchmarks), so you can re-run any number below. Every provider was called on August 4, 2026 under the same searcher configuration. Publication retrieval uses [LitSearch](https://github.com/princeton-nlp/LitSearch), an external academic dataset; we built the people and company sets because no public dataset covers entity-level retrieval. Benchmark What it tests Exa Tavily People Search R@1 Finding a named person's profile **75.5%** 40.5% Company Search R@1 Finding a company matching criteria **81.5%** 61.3% Company Search R@3 Same, anywhere in the top three **88.3%** 73.0% Publication Retrieval R@1 Finding a paper from a description **63.3%** 31.8% ### Evaluation methodology - Exa figures are the fast search type; instant and auto score differently. Tavily figures are its advanced depth. - Dataset sizes: 200 queries each for People and Company, 597 LitSearch queries for Publication. - Non-Exa providers were applied a consistent domain constraint; removing this reduced Tavily's performance. - Scoring handled by GPT-5.4 according to published repository rubrics. Company search is the closest result here, at 88.3% against 73.0% in the top three. The people and publication gaps are wider, and the reason is structural rather than a matter of quality: Exa operates specialized indexes for people, companies and publications, whereas Tavily relies on general web retrieval for these tasks. ## Best use cases for the Exa search API Exa is the optimal choice for vertical retrieval of people, companies, code, and publications, and for agent loops where the slowest call sets the budget. ### Exa vs Tavily on people and company search Exa maintains proprietary indexes for people and companies. A query like "Series B fintech companies in Singapore with 50-200 employees" retrieves structured records instead of keyword-matched web pages. Index highlights: - 1B+ profiles with 50M+ weekly updates. - Company search powered by custom embeddings, filterable by headcount, funding, and industry. - Full retrieval benchmarks available at [github.com/exa-labs/benchmarks](https://github.com/exa-labs/benchmarks). Metric Exa Tavily Perplexity People R@1 **75.5%** 40.5% 53.5% Company R@1 **81.5%** 61.3% 69.3% Company R@3 **88.3%** 73.0% 83.3% Measured on 200 people queries and 200 company queries, August 4, 2026. Exa figures are the fast search type. While general search APIs often return job boards for professional queries, Exa directly surfaces profiles. **How Tavily compares:** The Tavily API lacks a dedicated entity search product, forcing these queries through its general web index. ### Exa for academic and publication search Exa maintains a proprietary index of 350M+ academic works, prioritizing direct retrieval of original papers. This was measured on LitSearch, which requires identifying a paper from a description rather than a title. Provider R@1 **Exa** **63.3%** Tavily 31.8% 597 LitSearch queries, August 4, 2026. Exa figures are the fast search type. The grader computes rank-one recall, so no deeper rank is reported. **How Tavily compares:** No publication-specific category; general retrieval frequently surfaces secondary summaries rather than primary papers. ### Exa for token-efficient RAG Retrieval pipelines that send full page text to the model pay for tokens the query never needed, and that cost lands on your model bill rather than the search API's. Exa's highlights are selected against your query, so the same URL returns different passages for different questions. Raising a token limit returns more of the page, not the part of the page that answers the question. Fewer tokens is not only a cost saving. Every token the model reads is time before your agent can act, so a smaller, better-targeted context shortens each step of the loop. **How Tavily compares:** The extract endpoint returns static page content, requiring your LLM to handle filtering at your own token expense. ### Exa vs Tavily on search API latency For chained agent loops, tail latency is the primary constraint. Exa's tighter distribution ensures more predictable finish times for multi-step workflows. Tier p50 p90 p99 Exa Instant 235ms **263ms** **437ms** Tavily ultra-fast 245ms 334ms 576ms 333 calls per provider, run from independent AWS environments on July 29, 2026. While median speeds are comparable, Exa is 27% faster at p90 and 32% faster at p99. **How Tavily compares:** Tavily is the closest competitor on latency but exhibits greater variability in the tail. ### Exa for production filtering Production-grade agents require strict control over source domains to ensure security and relevance. - `includeDomains` and `excludeDomains` support up to 1,200 entries each. - `includeText` and `excludeText` for precise content matching. - Support for ISO 8601 ranges, `maxAgeHours`, and language filters. **How Tavily compares:** Offers date presets and country boosting but lacks deep category filtering and restricts domain lists to 300/150 entries. ## Best use cases for the Tavily API Tavily is a well-built platform with distinct operational advantages. ### Credit-based pricing on pay-as-you-go volume Tavily bills one credit for `basic` and two for `advanced` searches. This unified unit across search, extraction, and mapping simplifies financial modeling for some teams. **How Exa compares:** Exa uses request-based pricing ($7/1k) with separate lines for additional results or content types. Enterprise agreements with Exa are negotiated, so teams that need billing predictability at volume get it through the contract rather than through the pricing unit. ### A higher published request-rate ceiling Tavily publishes a 1,000 RPM limit on its standard tier, which may suit crawling-heavy workloads that prioritize volume over rank-one recall. **How Exa compares:** Exa publishes 10+ QPS with custom scaling available for enterprise accounts. ### Consolidation on Nebius For teams already utilizing the Nebius AI cloud, Tavily offers vendor consolidation as it integrates further into that ecosystem. **How Exa compares:** Exa remains an independent laboratory with a singular focus on AI search capabilities. ## Exa search API platform overview A unified semantic index powering Exa's API products. ### Exa Search API [Exa Search](https://exa.ai/products/search) provides the highest accuracy of any search API within its latency class. It processes natural language queries to deliver results optimized for token efficiency, complete with citations. Users can select from [six search types](https://exa.ai/docs/reference/search) spanning budgets from ~250ms to tens of seconds. This includes `instant` and `fast` for speed-sensitive loops, `auto` for adaptive scaling, and a range of [deep research modes](https://exa.ai/products/deep): `deep-lite`, `deep`, and `deep-reasoning`. ### Exa Agent API The [Exa Agent](https://exa.ai/products/agent) is a high-compute research primitive designed to synthesize information from the entire web. It automates multi-hop reasoning for complex workflows like exhaustive list-building and entity enrichment. Compute is governed by effort levels ranging from `low` to `xhigh`, alongside an `auto` mode that scales resources to meet the task requirements. ### Exa Contents API The [Contents API](https://exa.ai/docs/reference/get-contents) retrieves full-page text, summaries, and structured data according to a user-defined schema. It automatically resolves complex layouts, including PDFs and pages requiring JavaScript rendering. Query-dependent highlights reduce token consumption significantly; just 500 characters of highlights provide the same retrieval accuracy as the initial 8,000 characters of page content, all with sub-100ms latency. ### Exa Code API The [context endpoint](https://exa.ai/docs/reference/context), also called Exa Code, is tuned for coding agents and developer workflows. It searches repositories, documentation, and technical forums, returning results formatted as usable code context. A `tokensNum` parameter allows for precise token budgeting on each call. ### Exa Monitors API The [Monitors API](https://exa.ai/docs/reference/monitors/create-a-monitor) runs saved searches on a schedule and surfaces what changed since the last run. ## How to choose between the Exa and Tavily APIs If you need Choose Why People or company search Exa 75.5% vs 40.5% R@1 on people Academic retrieval Exa 63.3% vs 31.8% R@1 Lowest token overhead Exa Query-selected highlights instead of static page content Predictable tail latency Exa 32% faster at p99 Large domain lists Exa 1,200 each vs 300/150 Bundled raw content Tavily Raw text included in credit price Highest published RPM Tavily 1,000 req/min standard limit Teams building agents that search for specific entities rather than general pages will typically find Exa to be the stronger choice for quality and cost-per-answer. ## Switching from the Tavily API to Exa Migration is mostly a mapping exercise. Integrations with LiteLLM, Vercel AI SDK, LangChain, and [many other providers](https://exa.ai/integrations) are natively supported, simplifying the cutover. You'll want to: - Find equivalent parameters in the Exa API: - `max_results` becomes `num_results` (limit: 100). - Tavily's `include_domains` and `exclude_domains` become `includeDomains` and `excludeDomains`, with the limits rising from 300 and 150 to 1,200 each. - Replace date presets with specific start/end dates. - Map Tavily depth to Exa search types: - `ultra-fast` maps to Exa `instant`. - For other depths, use `auto` by default for reliable ~1s responses. - Optimize content retrieval: - Replace `include_raw_content` with `contents.highlights` to maximize token savings. ## Exa vs Tavily FAQ