BENCHMARKS

Top performance across benchmarks.

Exa Company Benchmark

Last updated: 08/30/26

ABOUT THE BENCHMARK

This benchmark measures how accurately web search APIs can answer dynamic questions about companies. Examples include funding, headcount, new product launches.

Precision @10
0.20.40.6
Exa
Parallel
Perplexity
Tavily
Serper
Linkup
Firecrawl
10002000300040005000
Average latency (p50 ms)

Exa Company Benchmark

Last updated: 08/30/26

ABOUT THE BENCHMARK

This benchmark measures how accurately web search APIs can answer dynamic questions about companies. Examples include funding, headcount, new product launches.

Seltz. Speed: 318ms. P@10: 0.56

Exa Company Benchmark

Last updated: 08/30/26

ABOUT THE BENCHMARK

This benchmark measures how accurately web search APIs can answer dynamic questions about companies. Examples include funding, headcount, new product launches.

Seltz. Speed: 318ms. P@10: 0.56

Benchmark details

Benchmark details

PUBLIC BENCHMARK

Exa Company Benchmark

ABOUT

Exa's open-source company benchmark measures 950 queries related to industry, geography, headcount and funding stage.

EVALUATION

It tests whether a search API can find companies by facts that keep changing. Providers score well here by retrieving rather than by recalling a company it already knows.

RESULT

More of the ten companies we return are ones you'd actually use, so there's less to throw away before the list is workable. The right company lands somewhere in those ten as often as with Exa. Each search takes about a third of a second, which is fast enough to run several inside a live request.

Company
Accuracy @P10
56%
Exa
54%
Parallel
51%
Perplexity
32%
Linkup
16%
Firecrawl
16%
Tavily
16%
Serper
15%

PUBLIC BENCHMARK

Exa People Benchmark

ABOUT

Exa's open source benchmark for people search. It covers 1,400 questions, each describing someone by what they do, how senior they are, and where they work.

EVALUATION

It tests whether an API can find the right person when you describe them instead of naming them, like a senior security lead in Berlin. The people in the set are real but not famous, so a system can't recall them and has to work from profile data that's current.

RESULT

Describe a person by role, level, and city, and you get the right profile about as often as with any provider tested. Each lookup takes roughly a third of a second, and the slow ones stay close to that, so a product checking six people in a row finishes in seconds rather than half a minute.

Company
Accuracy @P1
92%
Exa
92%
Parallel
85%
Perplexity
82%
Serper
52%
Tavily
50%
Linkup
28%
Firecrawl
5%

PUBLIC BENCHMARK

OpenBenchmarks

ABOUT

Openbenchmarks scores nine lookalike providers across 48 B2B seed companies and added us to the board themselves. Every number is reproducible from the published audit trail.

EVALUATION

It tests whether a provider can take one company and return others a buyer would seriously consider alongside it. Lists are scored at ten rows and at a hundred, because a shortlist someone works by hand and a long list an automated sequence runs through are two different jobs.

RESULT

Give us one of your best customers and ask for a hundred similar companies, and more of that hundred hold up than with anyone else on the board, so a sequence can run the full list rather than just the top of it. Per dollar spent, we return the most usable companies of any provider measured.

Company
Precision@100
Relevant / $1 at 100
63.4%
12,683
Extruct
61.1%
624
Ocean.io
59.5%
37
Parallel
56.9%
11,383
Exa
48.6%
501
Discolike
34.6%
66

Try the API today.

Grab an API key and get 20,000 search credits to start.

Fast, up-to-date web data, providing context-engineered web signals with sources for real-time AI reasoning.

Fast, up-to-date web data, providing context-engineered web signals with sources for real-time AI reasoning.

Fast, up-to-date web data, providing context-engineered web signals with sources for real-time AI reasoning.