Cheapest web search API for research agents: cost per research task
Ranked on the multi-hop search task of the web search benchmark. Each question combines three or four constraints.
Perplexity (perplexity) is cheapest at $0.665 LLM $ / task quality on this run.
Ranked by LLM $ / task quality: model cost per run divided by F1. Search API list prices per 1,000 queries are in the FAQ below; they are excluded from the sort because they dominate and vary by tier.
Most accurate · Fastest · Cheapest · Find companies by criteria · Search + fetch · Full benchmark
Ranking, sorted by llm $ / task quality
| Provider | Endpoint & configuration | LLM $ / task quality | Task quality (F1) | LLM $ / run |
|---|---|---|---|---|
| Perplexity | POST /searchsearch_context_size=low | $0.665 | 37.8% | $0.252 |
| Firecrawl | POST /v2/searchfirecrawl | $0.711 | 30.4% | $0.216 |
| Brave Search | GET /res/v1/web/searchbrave | $0.738 | 28.0% | $0.207 |
| TinyFish | GET api.search.tinyfish.aitinyfish | $0.784 | 26.6% | $0.209 |
| Nimble | POST /v2/searchsearch_depth=lite · full_content=false · focus=general | $0.825 | 24.1% | $0.199 |
| Parallel fast | POST /v1/search mode=fastparallel-fast | $1.110 | 38.0% | $0.422 |
| Exa deep | POST /search type=deepexa-deep | $1.119 | 45.4% | $0.508 |
| Parallel turbo | POST /v1/search mode=turboparallel-turbo | $1.125 | 34.7% | $0.391 |
| Parallel advanced | POST /v1/search mode=advancedparallel-advanced | $1.154 | 44.2% | $0.510 |
| Exa instant | POST /search type=instantexa-instant | $1.193 | 43.3% | $0.517 |
| Nimble | POST /v2/searchsearch_depth=standard · full_content=false · focus=general | $1.276 | 30.7% | $0.392 |
| Tavily basic | POST /search search_depth=basictavily-basic | $1.295 | 36.5% | $0.473 |
| Tavily advanced | POST /search search_depth=advancedtavily-advanced | $1.727 | 41.1% | $0.709 |
| Linkup fast | POST /v1/search depth=fastlinkup-fast | $1.881 | 41.1% | $0.774 |
| Linkup standard | POST /v1/search depth=standardlinkup-standard | $1.911 | 40.6% | $0.776 |
| Parallel basic | POST /v1/search mode=basicparallel-basic | $1.982 | 46.5% | $0.922 |
| You | POST /v1/searchextraction_mode=highlights · knowledge=core | $2.025 | 38.1% | $0.771 |
| You | POST /v1/searchextraction_mode=highlights | $2.119 | 38.6% | $0.818 |
| Seltz | POST /v1/search scope=companiesseltz-companies | $9.953 | 14.5% | $1.446 |
| SERP (RapidAPI) | GET google-search74.p.rapidapi.comserp | $18.570 | 0.4% | $0.069 |
LLM $ / run is model cost only; search API cost is excluded because it dominates. LLM $ / task quality divides that by F1 (task completion quality). Ranked by LLM $ / task quality.
Search plus fetch, same sort
| Provider | Endpoint & configuration | LLM $ / task quality | Task quality (F1) | LLM $ / run |
|---|---|---|---|---|
| Firecrawl | POST /v2/searchfirecrawl | $0.710 | 33.2% | $0.236 |
| TinyFish | GET api.search.tinyfish.aitinyfish | $0.715 | 30.2% | $0.216 |
| Brave Search | GET /res/v1/web/searchbrave | $0.730 | 29.4% | $0.215 |
| Nimble | POST /v2/searchsearch_depth=lite · full_content=false · focus=general · extract formats=[markdown] | $0.884 | 25.8% | $0.228 |
| Perplexity | POST /searchsearch_context_size=high | $0.921 | 46.6% | $0.429 |
| Exa deep | POST /search type=deepexa-deep | $1.022 | 48.2% | $0.492 |
| Parallel fast | POST /v1/search mode=fastparallel-fast | $1.044 | 39.3% | $0.410 |
| Parallel turbo | POST /v1/search mode=turboparallel-turbo | $1.068 | 36.0% | $0.385 |
| Exa instant | POST /search type=instantexa-instant | $1.161 | 44.9% | $0.521 |
| Parallel advanced | POST /v1/search mode=advancedparallel-advanced | $1.198 | 42.2% | $0.506 |
| Tavily basic | POST /search search_depth=basictavily-basic | $1.275 | 38.1% | $0.487 |
| Nimble | POST /v2/searchsearch_depth=standard · full_content=false · focus=general · extract formats=[markdown] | $1.315 | 31.3% | $0.412 |
| Tavily advanced | POST /search search_depth=advancedtavily-advanced | $1.528 | 41.0% | $0.627 |
| Linkup standard | POST /v1/search depth=standardlinkup-standard | $1.867 | 42.0% | $0.784 |
| Linkup fast | POST /v1/search depth=fastlinkup-fast | $1.909 | 39.9% | $0.761 |
| You | POST /v1/searchextraction_mode=highlights | $1.987 | 38.5% | $0.766 |
| You | POST /v1/searchextraction_mode=highlights · knowledge=core | $2.045 | 38.5% | $0.788 |
| Parallel basic | POST /v1/search mode=basicparallel-basic | $2.238 | 42.3% | $0.947 |
| Seltz | POST /v1/search scope=companiesseltz-companies | $9.015 | 16.3% | $1.467 |
| SERP (RapidAPI) | GET google-search74.p.rapidapi.comserp | — | 0.0% | $0.069 |
LLM $ / run is model cost only; search API cost is excluded because it dominates. LLM $ / task quality divides that by F1 (task completion quality). Ranked by LLM $ / task quality.
Common questions
What is the cheapest web search API for research agents?
Perplexity (perplexity) is cheapest at $0.665 LLM $ / task quality (search-only): model cost per run divided by F1. Search API spend is excluded from the sort because it dominates and varies by tier; list prices are below.
What do these web search APIs cost per 1,000 queries?
Public list prices for the configurations measured: Perplexity (perplexity) $5 per 1,000 queries; Firecrawl (firecrawl) $5 per 1,000 queries; Brave Search (brave) $5 per 1,000 queries; TinyFish (tinyfish) $0 per 1,000 queries; Nimble (nimble-lite) $1.100 per 1,000 queries; Parallel fast (parallel-fast) $1 per 1,000 queries; Exa deep (exa-deep) $12 per 1,000 queries; Parallel turbo (parallel-turbo) $1 per 1,000 queries; Parallel advanced (parallel-advanced) $5 per 1,000 queries; Exa instant (exa-instant) $7 per 1,000 queries; Nimble (nimble-standard) $5 per 1,000 queries; Tavily basic (tavily-basic) $8 per 1,000 queries; Tavily advanced (tavily-advanced) $16 per 1,000 queries; Linkup fast (linkup-fast) $5 per 1,000 queries; Linkup standard (linkup-standard) $5 per 1,000 queries; Parallel basic (parallel-basic) $5 per 1,000 queries; You (you-highlights-core) $5 per 1,000 queries; You (you-highlights) $5 per 1,000 queries; Seltz (seltz-companies) $5 per 1,000 queries; SERP (RapidAPI) (serp) $3 per 1,000 queries. List price is what the vendor publishes for the tier used here; it excludes free allowances and volume discounts, and it is not the cost of a research task, which depends on how many searches the agent makes.