People search benchmark 2026: Crustdata vs Exa vs Parallel
Last updated October 2026
Crustdata returned the most accurate people-search results of the three providers tested, with a 95.5% strict pass rate across 100 queries, against 83.7% for both Exa and Parallel. Strict pass rate is the share of returned people who satisfy every clause of the query. Crustdata and Exa both answer in under four seconds, while Parallel's deep-research API takes about 11 minutes per query. Every number here traces to a raw API response.
tldr:
- Crustdata had the highest strict pass rate, 95.5% across 100 people-search queries, against 83.7% for Exa and 83.7% for Parallel.
- The gap is role and topic. On location, seniority and tenure the three are even, all between 95% and 100%.
- Crustdata (3.7s) and Exa (3.8s) return in seconds. Parallel's async task API takes 11m 27s per query.
Which people-search provider is most accurate?
Crustdata is the most accurate, at a 95.5% strict pass rate, ahead of Exa and Parallel at 83.7% each.
| Provider | Strict pass rate | R@1 | Usable people per query |
|---|---|---|---|
| Crustdata | 95.5% | 95% | 9.55 / 10 |
| Exa | 83.7% | 88% | 8.37 / 10 |
| Parallel | 83.7% | 90% | 8.37 / 10 |
Each query asked for ten people. Crustdata returned 9.55 who satisfied every clause, against 8.37 for the other two. Exa and Parallel land on the same 83.7% by different routes, with Parallel a little stronger on the top result at 90% R@1 to Exa's 88%.
What makes Crustdata more accurate?
The three providers are close on most of the query. Crustdata separates on two clauses, role and topic.
| Clause | Crustdata | Exa | Parallel |
|---|---|---|---|
| Role | 97.2 | 88.4 | 86.1 |
| Location | 99 | 95.2 | 99 |
| Topic | 66.7 | 43.3 | 40 |
| Seniority | 100 | 96.7 | 100 |
| Tenure | 100 | 100 | 100 |
Location, seniority and tenure are close to solved for everyone. Tenure is a straight 100 across all three. The separation is role, where Crustdata clears 97% against the high 80s, and topic, which is the hardest clause for all three. Even Crustdata only gets topic right two times in three, and Exa and Parallel land closer to four in ten.
How fast is each provider?
Crustdata and Exa are interactive, at 3.7s and 3.8s per query. Parallel uses an async task API made for deep research, which returns in 11m 27s.
| Provider | Median time per query |
|---|---|
| Crustdata | 3.7s |
| Exa | 3.8s |
| Parallel | 11m 27s |
For a product that searches while a user waits, seconds matter. For a background job that can take minutes, Parallel's model is a reasonable trade. These are different categories, so read the latency next to what you are building.
How was this benchmark measured?
Each query was broken into typed clauses, role, location, topic, seniority and tenure. A returned person passes strict only if every clause passes.
- 100 people-search queries, 10 people requested per query.
- An LLM judge grades each clause against each returned person. Numeric clauses like tenure are checked directly against the record.
- Strict pass rate is the share of returned people who satisfy every clause.
- You can repeat the same eval on your own accounts and keys.
Grading is done by an LLM judge, not scored against any provider's own answers.
The bottom line on each provider
- Crustdata - highest strict pass rate at 95.5%, with the lead coming from role and topic, and interactive latency at 3.7s. The fit when you need accuracy on multi-clause queries in real time.
- Exa - 83.7% strict pass, interactive at 3.8s, and even with Crustdata on seniority and tenure. If your queries lean on role and topic, the gap shows there.
- Parallel - 83.7% strict pass and tied with the field on location, seniority and tenure. Its async deep-research API fits background work rather than live search, at about 11m 27s per query.
How to benchmark people-search providers yourself
This benchmark scores a general query set, not the day-to-day searches your team depends on. Take 100 of your own queries, send them to all three providers with your own keys, and grade the results clause by clause on what your users actually ask for.
FAQ
Which people-search API is most accurate in 2026? Crustdata, at a 95.5% strict pass rate across 100 queries, ahead of Exa and Parallel at 83.7% each.
Is Crustdata or Exa better for people search? Crustdata returns a higher strict pass rate, 95.5% against 83.7%, with the difference concentrated in role and topic accuracy. The two are even on location, seniority and tenure, and both answer in under four seconds.
What is strict pass rate? The share of returned people who satisfy every clause of the query. A person counts only when role, location, topic, seniority and tenure all pass.
Why is topic accuracy low for every provider? Topic is the hardest clause to match. It asks whether a person has actually worked on a subject, not whether a keyword appears. Crustdata leads at 66.7%, with Exa at 43.3% and Parallel at 40%.
How fast is each provider? Crustdata returns in 3.7s and Exa in 3.8s. Parallel's async task API returns in 11m 27s, since it is built for deep research rather than live search.