Exa Lite search API

The switch is one line

Your code keeps its body, type: "lite" included, and its parser. The address changes, and the key.

# Before: Exa Lite, $4 per 1,000
curl -X POST https://api.exa.ai/search \
  -H "x-api-key: $EXA_API_KEY" -H "Content-Type: application/json" \
  -d '{"query": "latest chip export rules", "type": "lite", "numResults": 5, "contents": {"text": true}}'

# After: SpecStream, $1 per 1,000, each page read live
curl -X POST https://api.specstream.ai/exa/search \
  -H "x-api-key: $SPECSTREAM_API_KEY" -H "Content-Type: application/json" \
  -d '{"query": "latest chip export rules", "type": "lite", "numResults": 5, "contents": {"text": true}}'

The reply has Exa's shape: requestId, results[] with id, url, title and publishedDate, text and highlights when you ask for them, searchTime and costDollars.total. There is no field to map.

Pages read live, at a quarter of the Lite price

Lite gets its price from its cache. It never fetches a page live, and a result with no cached copy is dropped when you ask for its text. Exa's docs send a request that needs a fresh page to auto, at $7 per 1,000. SpecStream reads each page at the moment of your call, for $1 per 1,000 searches of up to 10 results, whichever type you send except deep. A result whose page is not read in time keeps its snippet in text, so no result is dropped.

Your body is accepted as it is

Lite answers 400 INVALID_REQUEST_BODY to outputSchema, contents.summary, contents.maxAgeHours, contents.livecrawl and the deprecated includeText and excludeText. SpecStream's route reads includeText and excludeText as exact phrases, and accepts contents.maxAgeHours and contents.livecrawl without an error, since every page is read live. Code written for auto runs at the same $1 as code written for lite.

The most accurate web search API in the world

On SimpleQA Verified the SpecStream Search API scores 97.9 F-score on all 1,000 official questions, 978 of them correct. On WideSearch it scores 64.0 item F1 and 43.4 row F1 on all 200 official tasks. The Extract API scores 0.908 F1 on WCXB. Every score and its method is on the benchmarks page.

How fast the route answers

The fast search behind the Exa route had a median of 0.82 seconds at one search a second and 1.07 seconds at 20 searches a second, measured on production on November 9, 2026, with 718 of 720 searches answered. There is no limit per key and no daily limit.

Pages, documents and rows on the same key

POST /exa/contents answers Exa's Contents call, and exa.get_contents(urls) works with the same base_url: up to 20 pages a call, each at the Extract API's $0.80 per 1,000 pages, PDF, Word and Excel included. When you need every match on the web in place of a results page, Deep Search returns structured rows at 5¢, 15¢ or $1 a row, each value with its source URL and read time, under a maximum you approve before it runs.

Exa auto

Exa's default type costs $7 per 1,000 requests and fetches a page when its cache has none. It is the type Exa's docs name for any request that needs a parameter lite refuses. SpecStream answers an auto body on the same route for $1 per 1,000.

Parallel Search API

Search starts at $1 per 1,000 requests, with Parallel's own request body and JSON. SpecStream answers Parallel's call too, at https://api.specstream.ai/parallel/v1/search, on the same key as the Exa route.

Perplexity Search API

Ranked results as JSON at $5 per 1,000 requests, and $1 with Fast Search. SpecStream answers Perplexity's POST /search at https://api.specstream.ai/perplexity, with every page read live.

Brave Search API

A search API on Brave's own index at $5 per 1,000 requests, with $5 in free credits every month. SpecStream answers Brave's /res/v1/web/search call on the same path, at $1 per 1,000 searches of up to 10 results.

Tavily

Billed in credits at $0.008 each pay as you go, and a basic search is 1 credit, so $8 per 1,000. SpecStream answers Tavily's call at https://api.specstream.ai/tavily/search for $1 per 1,000.