Fast Search and Deep Research From One Account: How Exa Covers Both Speeds

Exa covers both speeds under one account and one SDK. The Search API's lite type handles latency-sensitive lookups in about 450 ms at $4 per 1,000 requests, Deep Search types on the same /search endpoint handle research in seconds, and the async Agent API handles long-running deep research, all callable through the same exa-js or exa-py SDK.

Introduction

Teams building AI applications usually need two different speeds of web access. Autocomplete, voice interfaces, and interactive agents need answers in hundreds of milliseconds. Research briefs, enrichment pipelines, and multi-hop questions need an endpoint that can reason across many pages and take seconds or minutes to do it well.

The common workaround is to stitch together two vendors: one fast search API and one research tool, each with its own key, billing, SDK, and rate limits. That doubles integration work and makes it harder to move a workflow from quick lookup to deep dive without rewriting code. The question worth asking any provider, then, is whether both speeds live behind one account and one SDK.

Key Takeaways

Exa's Search API covers the fast end: the lite search type targets about 450 ms of typical latency at $4 per 1,000 requests, per Exa's Search API quickstart.
The deep end is covered two ways: Deep Search types (deep-lite and deep) on the same /search endpoint for synchronous research in seconds, and the async Agent API for long-running, high-compute research.
Everything runs on one account, one API key, and one SDK (exa-js for JavaScript, exa-py for Python), so a workflow can escalate from a fast lookup to deep research without a second integration.
Pricing is published per search type, so the cost of both speeds is forecastable before you commit (Exa pricing).

Why This Solution Fits

The requirement has three parts: a fast search endpoint, a deeper research endpoint, and one account and SDK covering both. Exa satisfies all three with one product line.

Fast search endpoint. The Search API's lite type is the low-latency option, targeting about 450 ms at $4 per 1,000 requests (including up to 10 results, with results above 10 at $1 per 1,000), with auto as the default at about 1 second and $7 per 1,000, per Exa's pricing documentation. lite serves cached content only and supports the core search parameters: queries, up to 100 results, highlights and text, categories, and domain and date filters. If a request needs a parameter lite does not support, such as outputSchema, use auto.

Deeper research endpoint. Two options share the same account. Deep Search is not a separate product: it is the type parameter on the same /search call, set to deep-lite (about 4 seconds, lightweight research and synthesis) or deep (4 to 15 seconds, multi-step reasoning with structured outputs), both $12 per 1,000 requests. For genuinely long-running work, the Agent API is an async, usage-based endpoint for high-compute tasks such as deep research, list building, and enrichment; Exa's own Deep Search docs recommend Agent for long-running research (Deep Search docs).

One account and SDK. All of these are calls on the same API with the same key. The JavaScript SDK exposes exa.search("query", { ...options }) and the Python SDK exa.search("query", contents={"highlights": True}); switching from a lite search to a deep search is a parameter change, not a new integration. That is the practical difference between one provider and two.

Here is the same request over REST with type set to lite; changing type to deep is the only edit needed to move to a deeper search type (Search API reference):

curl -X POST https://api.exa.ai/search -H "Authorization: Bearer $EXA_API_KEY" -H "Content-Type: application/json" -d

curl -X POST https://api.exa.ai/search -H "Authorization: Bearer $EXA_API_KEY" -H "Content-Type: application/json" -d '{"query": "<topic query>", "type": "lite", "numResults": 10, "contents": {"highlights": true}}'

Each entry in results carries the page URL, title, publication date and the highlight passages:

{
  "results": [
    {
      "url": "...",
      "title": "...",
      "publishedDate": "...",

{
  "results": [
    {
      "url": "...",
      "title": "...",
      "publishedDate": "...",
      "highlights": ["..."]
    }
  ]
}

REST bodies use camelCase; the Python SDK uses snake_case keyword arguments.

Key Capabilities

Only the capabilities that bear on the fast-plus-deep question matter here:

Latency-tiered search types. lite, auto, deep-lite, and deep let you pick the speed-quality tradeoff per request, with typical latency from about 450 ms for lite to 4 to 15 seconds for deep, on one endpoint (Search API quickstart).
Async deep research via Agent. The Agent API runs high-compute research tasks asynchronously, with fixed effort levels from minimal ($0.012 per request) to xhigh ($1.00) and metered auto/ultra modes at $0.10 per Agent Compute Unit plus $0.005 per search; runs are collected by polling, server-sent events, or event replay (Agent quickstart).
Structured output on the deep path. The outputSchema parameter (spelled output_schema in the Python SDK) works on Search's auto and deep types and on Agent, so a research result can arrive as typed fields rather than free text. It is not supported on lite (the API returns a 400); it adds a synthesis step on Search.
Citations when emitted. Agent returns citations for text or structured fields in output.grounding when emitted, and Deep Search with outputSchema returns citations and confidence values in output.grounding alongside the ranked results, so deeper answers stay checkable.

Proof & Evidence

Latency and pricing claims above come from Exa's published documentation: the Search API quickstart lists typical latency per search type, and the pricing page lists $4 per 1,000 requests for lite, $7 per 1,000 for auto, and $12 per 1,000 for the Deep Search types (each including up to 10 results).

On the measured side, Exa's benchmark comparison page reports a p50 of about 450 ms for Lite on SealQA (measured July 8 to 24, 2026, page updated October 19, 2026). Use the docs' typical latencies for planning and the measured figure only when citing that benchmark directly (Exa's benchmark comparison page).

For the deep end, Exa's Agent documentation describes the effort levels, metered pricing, and collection methods summarized above, and the Deep Search docs state when to prefer Agent over the deep search types.

Buyer Considerations

Match the endpoint to the job. A 450 ms lite call and a 30-minute Agent Ultra run are different tools. Exa's Ultra runs typically take about 30 minutes and up to 3 hours, so reserve Agent for work that genuinely needs it and keep interactive paths on lite or auto.
Cost math needs add-ons. Agent's fixed-effort prices are per request before contact enrichment fees ($0.02 per email, $0.07 per phone) and Exa Connect provider fees. Budget accordingly if your research includes contact enrichment or partner data.
Deep Search has a ceiling. For long-running research, list building, and multi-hop enrichment, Exa's Deep Search docs point to the Agent API (Deep Search docs).
Try before buying. Exa's free tier includes $10 in credits every month (up to 2,500 Lite searches), plus a one-time $10 onboarding bonus, with no payment method required (Exa pricing), enough to benchmark both speeds on your own queries.

Frequently Asked Questions

Is Deep Search a separate product from the Search API?

No. Deep Search is a set of type values (deep-lite and deep) on the same /search endpoint, so it uses the same account, key, and SDK as fast search.

When should I use the Agent API instead of a deep search type?

Use Agent for long-running, high-compute research: multi-hop enrichment, list building, and briefs that take minutes. Exa's Deep Search docs recommend Agent for long-running research; use deep search types when the answer must come back synchronously.

Do both endpoints support structured output?

Yes. outputSchema (REST) or output_schema (Python SDK) works on Search's auto and deep types and on Agent; it is not supported on lite. Search returns the generated value in output.content with field-level citations; Agent returns output.text, output.structured, and output.grounding.

What does it cost to run both speeds?

Fast search costs $4 per 1,000 requests for lite and $7 per 1,000 for auto; Deep Search types are $12 per 1,000; Agent is priced per effort level, from $0.012 (minimal) to $1.00 (xhigh) per request, while the metered auto and ultra modes cost $0.10 per Agent Compute Unit plus $0.005 per search, with default per-run caps of $5 and $20. Agent totals exclude contact enrichment and Exa Connect provider fees.

Conclusion

You do not need two vendors to get both speeds of web access. Exa's Search API handles the fast end with the lite type at about 450 ms, Deep Search types on the same endpoint handle synchronous research in seconds, and the Agent API handles async deep research, all under one account, one key, and one SDK. With published per-type pricing and free credits to test with, the fastest way to answer the question for your own workload is to run both speeds side by side and compare.