Search API with the Best Recall, 2026
Search API with the Best Recall: Parallel basic leads 12 search APIs at 34.4% recall on 45 company search tasks, independently benchmarked in 2026.
What this page compares. This page ranks measured results from 2 boards: multi-turn company search (search only) and multi-turn company search (search and page fetch).
How to read the result. The ranking uses recall, the share of correct companies found. Precision and F1 stay beside it, since recall alone rewards long lists.
Multi-turn company search, search only, ranked by recall
Every row ran the same 45 questions. Provider names link to the official docs for the configuration measured.
| Rank | Provider | F1 | Precision | Recall | Median task time | Median cost / task | Median task tokens | List price |
|---|---|---|---|---|---|---|---|---|
| 1 | Parallel basicparallel-basic | 46.5% | 88.7% | 34.4% | 67.6 s | $1.12 | 303,467 | $5.00 / 1k queries |
| 2 | Exa deepexa-deep | 45.4% | 83.2% | 33.7% | 89.5 s | $0.72 | 153,061 | $12.00 / 1k queries |
| 3 | Exa instantexa-instant | 43.3% | 82.6% | 32.2% | 49.8 s | $0.65 | 152,622 | $7.00 / 1k queries |
| 4 | Parallel advancedparallel-advanced | 44.2% | 87.6% | 32.0% | 83.4 s | $0.62 | 159,733 | $5.00 / 1k queries |
| 5 | Linkup fastlinkup-fast | 41.1% | 82.8% | 30.3% | 64.2 s | $0.92 | 259,642 | $5.00 / 1k queries |
| 6 | Tavily advancedtavily-advanced | 41.1% | 83.7% | 29.8% | 92.2 s | $1.03 | 237,706 | $16.00 / 1k queries |
| 7 | Linkup standardlinkup-standard | 40.6% | 84.0% | 29.7% | 72.3 s | $0.94 | 249,382 | $5.00 / 1k queries |
| 8 | Youextraction_mode=highlights | 38.6% | 79.3% | 28.3% | 48.2 s | $0.97 | 264,016 | $5.00 / 1k queries |
| 9 | Parallel fastparallel-fast | 38.0% | 79.9% | 27.5% | 53.9 s | $0.46 | 119,953 | $1.00 / 1k queries |
| 10 | Youextraction_mode=highlights · knowledge=core | 38.1% | 80.9% | 27.4% | 47.6 s | $0.90 | 244,245 | $5.00 / 1k queries |
| 11 | Perplexitysearch_context_size=low | 37.8% | 79.3% | 26.8% | 48.9 s | $0.33 | 66,928 | $5.00 / 1k queries |
| 12 | Tavily basictavily-basic | 36.5% | 82.5% | 25.7% | 63.0 s | $0.61 | 142,692 | $8.00 / 1k queries |
| 13 | Parallel turboparallel-turbo | 34.7% | 80.0% | 24.8% | 46.4 s | $0.42 | 110,844 | $1.00 / 1k queries |
| 14 | Nimblesearch_depth=standard · full_content=false · focus=general | 30.7% | 70.3% | 21.3% | 53.0 s | $0.47 | 109,000 | $5.00 / 1k queries |
| 15 | Firecrawlfirecrawl | 30.4% | 77.3% | 20.7% | 75.0 s | $0.28 | 50,596 | $5.00 / 1k queries |
| 16 | Brave Searchbrave | 28.0% | 66.9% | 19.3% | 43.5 s | $0.27 | 49,005 | $5.00 / 1k queries |
| 17 | TinyFishtinyfish | 26.6% | 64.7% | 17.9% | 64.2 s | $0.21 | 51,091 | $0 / 1k queries |
| 18 | Nimblesearch_depth=lite · full_content=false · focus=general | 24.1% | 65.2% | 16.1% | 67.2 s | $0.22 | 50,682 | $1.10 / 1k queries |
| 19 | Seltzseltz-companies | 14.5% | 40.0% | 9.4% | 55.2 s | $1.75 | 500,633 | $5.00 / 1k queries |
| 20 | SERP (RapidAPI)serp | 0.4% | 0.7% | 0.3% | 31.4 s | $0.10 | 10,890 | $3.00 / 1k queries |
Multi-turn company search, search and page fetch, ranked by recall
| Rank | Provider | F1 | Precision | Recall | Median task time | Median cost / task | Median task tokens | List price |
|---|---|---|---|---|---|---|---|---|
| 1 | Exa deepexa-deep | 48.2% | 89.4% | 36.0% | 95.8 s | $0.68 | 147,079 | $12.00 / 1k queries |
| 2 | Perplexitysearch_context_size=high | 46.6% | 87.7% | 34.7% | 53.9 s | $0.50 | 121,239 | $5.00 / 1k queries |
| 3 | Exa instantexa-instant | 44.9% | 85.9% | 33.5% | 52.5 s | $0.65 | 156,796 | $7.00 / 1k queries |
| 4 | Parallel basicparallel-basic | 42.3% | 81.3% | 31.3% | 68.6 s | $1.09 | 304,826 | $5.00 / 1k queries |
| 5 | Linkup standardlinkup-standard | 42.0% | 90.7% | 30.5% | 81.0 s | $0.91 | 254,078 | $5.00 / 1k queries |
| 6 | Parallel advancedparallel-advanced | 42.2% | 87.6% | 30.1% | 80.9 s | $0.60 | 152,116 | $5.00 / 1k queries |
| 7 | Tavily advancedtavily-advanced | 41.0% | 89.4% | 29.1% | 92.6 s | $0.90 | 208,974 | $16.00 / 1k queries |
| 8 | Linkup fastlinkup-fast | 39.9% | 85.3% | 28.6% | 68.0 s | $0.90 | 246,395 | $5.00 / 1k queries |
| 9 | Youextraction_mode=highlights · knowledge=core | 38.5% | 78.2% | 28.3% | 48.5 s | $0.89 | 250,679 | $5.00 / 1k queries |
| 10 | Parallel fastparallel-fast | 39.3% | 82.3% | 28.2% | 55.1 s | $0.44 | 119,278 | $1.00 / 1k queries |
| 11 | Tavily basictavily-basic | 38.1% | 81.2% | 27.8% | 62.0 s | $0.61 | 141,892 | $8.00 / 1k queries |
| 12 | Youextraction_mode=highlights | 38.5% | 85.2% | 26.9% | 45.6 s | $0.86 | 241,834 | $5.00 / 1k queries |
| 13 | Parallel turboparallel-turbo | 36.0% | 83.6% | 25.0% | 48.1 s | $0.41 | 108,358 | $1.00 / 1k queries |
| 14 | Firecrawlfirecrawl | 33.2% | 83.3% | 22.7% | 82.1 s | $0.29 | 55,069 | $5.00 / 1k queries |
| 15 | Nimblesearch_depth=standard · full_content=false · focus=general · extract formats=[markdown] | 31.3% | 73.2% | 21.7% | 59.2 s | — | 117,345 | $5.00 / 1k queries |
| 16 | TinyFishtinyfish | 30.2% | 70.9% | 20.9% | 63.4 s | $0.22 | 53,654 | $0 / 1k queries |
| 17 | Brave Searchbrave | 29.4% | 73.5% | 20.4% | 45.1 s | $0.28 | 54,482 | $5.00 / 1k queries |
| 18 | Nimblesearch_depth=lite · full_content=false · focus=general · extract formats=[markdown] | 25.8% | 67.2% | 17.7% | 72.4 s | — | 55,416 | $1.10 / 1k queries |
| 19 | Seltzseltz-companies | 16.3% | 49.5% | 10.2% | 60.1 s | $1.74 | 502,391 | $5.00 / 1k queries |
| 20 | SERP (RapidAPI)serp | 0.0% | 0.0% | 0.0% | 33.4 s | $0.10 | 11,051 | $3.00 / 1k queries |
How the ranking was measured
- A research agent searches over several turns to list every company that matches a set of constraints, on the same questions for every API.
- Answers are scored against a fixed list of correct companies: precision, recall and F1.
- Search-only rows combine search results; search-and-fetch rows can also read pages.
- Median task time, task tokens and median cost per task are reported beside the score.
- The runner and published data are open in openbenchmarks-labs/multi-turn-company-search ↗ and openbenchmarks/OB-Company-Websearch ↗. The full methodology is on the Multi-Turn Company Search Benchmark →
Questions answered by this comparison
Which provider leads search API with the Best Recall?
Parallel basic leads this table at 34.4% recall, on 45 company search tasks.
How was this measured?
A research agent searches over several turns to list every company that matches a set of constraints, on the same questions for every API. Answers are scored against a fixed list of correct companies: precision, recall and F1. Search-only rows combine search results; search-and-fetch rows can also read pages. Median task time, task tokens and median cost per task are reported beside the score.
How should speed and cost be read beside the score?
Parallel basic recorded 67.6 s median task time and $1.12 median cost per task. The most accurate configuration is rarely the fastest or the cheapest, so each is reported in its own column.
Do vendors pay to be ranked?
No. Inclusion and rank do not depend on vendor payments. Every API used the same questions, answering model and scoring.
What does this page not test?
This page ranks measured results from 2 boards: multi-turn company search (search only) and multi-turn company search (search and page fetch).
Related pages
- Multi-Turn Company Search Benchmark: full leaderboard and methodology →
- Fastest Web Search API for Company Search
- Fastest Search API for Company Search
- Most Accurate Web Search API for Company Search
- Most Accurate Search API for Company Search
- Most Accurate Web Search API for Multi-Turn Company Search
- Fastest Web Search API for Multi-Turn Company Search
- Every multi-turn company search comparison →