Fastest search
348ms
Parallel turbo
71.3% accuracy · $1.00 per 1,000 queries.
independent benchmark · same questions · search latency
Fastest Web Search API for AI (2026): Latency Comparison & Benchmark
Compare web search APIs from 11 providers, including Exa, Firecrawl, Parallel, Tavily and Brave, for AI, LLM apps, agents and RAG. Fastest search: Parallel turbo 348ms average, 71.3% accuracy and $1.00 per 1,000 queries. Best search accuracy: Exa fast 99.3% at 652ms. Cheapest listed search: TinyFish $0.00 per 1,000 queries within published limits. Independent latency comparison and benchmark on 300 questions, with official docs and pricing.
Compared on the same 300 search questions. Last measured 12 September 2026. The main ranking uses average search response time.
Best search accuracy
99.3%
Exa fast
652ms average search time · $7.00 per 1,000 queries.
Cheapest listed search
$0.00
TinyFish · per 1,000 queries
Free search; 30 requests/minute and 500/hour. 2.62s average · 92.0% accuracy.
Web Search API Latency Comparison: Speed, Accuracy and Cost
Ranked by average search response time, fastest first. Median (p50) shows the middle response time. Compare accuracy and price for each exact mode to choose a fast search API for AI applications.
| API / mode | Average search time | Median (p50) | Accuracy | Price / 1,000 queries | Official docs & pricing |
|---|---|---|---|---|---|
| Parallel turbomode=turbo | 348ms | 315ms | 71.3% | $1.00Per-request price for up to 10 results; extra results/excerpts add cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Exa instanttype=instant | 398ms | 386ms | 97.7% | $7.00Per-request price for up to 10 results; extra results and summaries add cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Firecrawl | 510ms | 471ms | 95.3% | $5.00Standard-plan top-up rate: 2 credits/10 results. Subscription required; see the linked pricing comparison.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Brave Search llm-contextcount=10 | 601ms | 590ms | 94.0% | $5.00Prepaid Search plan; LLM Context included. Before monthly free credits.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| You highlightsextraction_mode=highlights | 628ms | 614ms | 90.7% | $5.00Web Search price, including highlights; live page extraction adds cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Brave Search webcount=10 · result_filter=web | 630ms | 613ms | 93.3% | $5.00Prepaid Search plan; LLM Context included. Before monthly free credits.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Exa fasttype=fast | 652ms | 569ms | 99.3% | $7.00Per-request price for up to 10 results; extra results and summaries add cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| SERPlimit=10 | 751ms | 578ms | 96.0% | $3.00Pro overage $0.003/requestRecorded 1 September 2026; confirm current rate with provider. | Official docs ↗Official pricing ↗ |
| Nimble standardsearch_depth=standard · full_content=false · focus=general | 861ms | 822ms | 93.0% | $5.00search_depth=standard · full_content=falseRecorded 1 September 2026; confirm current rate with provider. | Official docs ↗Official pricing ↗ |
| You highlights coreextraction_mode=highlights · knowledge=core | 889ms | 815ms | 92.0% | $5.00Web Search price, including highlights; live page extraction adds cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Parallel fastmode=fast | 942ms | 861ms | 86.0% | $1.00Per-request price for up to 10 results; extra results/excerpts add cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Perplexity lowsearch_context_size=low | 1.38s | 1.60s | 97.3% | $5.00Search API, POST /search; one query per request in this comparison.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Linkup fastdepth=fast · outputType=searchResults | 1.57s | 1.17s | 96.7% | $5.00Prepaid API-key pricing with outputType=searchResults.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Parallel basicmode=basic | 1.68s | 1.78s | 93.3% | $5.00Per-request price for up to 10 results; extra results/excerpts add cost.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Tavily basicsearch_depth=basic | 1.88s | 1.69s | 87.7% | $8.001 credits/search at $0.008/credit, pay as you go.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Linkup standarddepth=standard · outputType=searchResults | 2.55s | 2.10s | 92.0% | $5.00Prepaid API-key pricing with outputType=searchResults.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| TinyFishfree · 30 req/min cap | 2.62s | 2.27s | 92.0% | $0.00Free search; 30 requests/minute and 500/hour.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Nimble litesearch_depth=lite · full_content=false · focus=general | 3.15s | 1.39s | 75.3% | $1.10search_depth=lite · full_content=falsePricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
| Tavily advancedsearch_depth=advanced | 4.29s | 4.23s | 93.0% | $16.002 credits/search at $0.008/credit, pay as you go.Pricing checked 19 September 2026. | Official docs ↗Official pricing ↗ |
Compare search pricing, free limits and cost per 1,000 queries →
Parallel Turbo vs Exa Instant: Fast Search Modes Compared
Parallel turbo
Parallel turbo: 348ms average search time, 71.3% accuracy and $1.00 per 1,000 queries.
Median response time: 315ms. Use this row to assess the fastest measured response against the evidence accuracy your application needs.
Official docs ↗Exa instant
Exa instant: 398ms average search time, 97.7% accuracy and $7.00 per 1,000 queries.
Median response time: 386ms. Compare this speed and accuracy combination with Exa fast in the main table.
Official docs ↗Fastest with at least 95% measured accuracy: Exa instant at 398ms and 97.7%. This threshold is a practical shortlist filter on the same results.
Tavily, Brave and Firecrawl Search Latency
Firecrawl: 510ms average search time, 95.3% accuracy and $5.00 per 1,000 queries. Brave Search llm-context: 601ms average search time, 94.0% accuracy and $5.00 per 1,000 queries. Brave Search web: 630ms average search time, 93.3% accuracy and $5.00 per 1,000 queries. Tavily basic: 1.88s average search time, 87.7% accuracy and $8.00 per 1,000 queries. Tavily advanced: 4.29s average search time, 93.0% accuracy and $16.00 per 1,000 queries. Firecrawl’s listed rate is for Standard-plan top-ups and requires a subscription; Tavily uses its pay-as-you-go credit rate.
How We Benchmark Web Search API Latency
Each API receives the same 300 company-news questions, with one search per question and up to 10 results. Search time measures the API request; accuracy is the share of questions answered correctly from its returned evidence, using the same answer model and judge. The code and public dataset sample are available, with a held-out scoring set.
Mean, p50 and tail latency
Mean latency is the average response time across measured queries. Median, or p50, is the middle response time. P95 describes the slowest 5% boundary and helps size a production timeout. This page ranks by measured mean and shows median separately; collect p95 and p99 from your application’s query and concurrency profile when choosing production settings.
Compare like-for-like measurements
Latency depends on the query set, deployment region, result count, returned content, cache state, concurrency and search mode. Compare the same statistic under matching conditions: a vendor p50 and a benchmark mean summarize different things. This table uses one benchmark’s measurements, with endpoint settings and official docs linked for each mode.
Open source code ↗ · Public dataset sample ↗ · Full search benchmark →
Supporting Developer Search Results
100 developer questions. Compare average search time with median full task time, which also includes model work and repeated tool calls. Tables are ordered by average search time; accuracy is measured as task completion.
Developer search: latency, accuracy and cost
| Rank | Vendor | Endpoint & configuration | Official docs | Task completion | Median task time | Median task cost | Avg search time | Median task tokens | API list price |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Parallel turbo | POST /v1/search mode=turbo | Official docs ↗ | 64.7 ± 2.1 | 19s | $0.059$0.005 search$0.054 token cost | 333ms | 14,130 | $0.001 / search |
| 2 | Exa instant | POST /search type=instant | Official docs ↗ | 61.3 ± 2.9 | 21s | $0.105$0.035 search$0.070 token cost | 447ms | 22,423 | $0.007 / search |
| 3 | Brave | POST /res/v1/llm/context | Official docs ↗ | 38.0 ± 2.6 | 20s | $0.082$0.025 search$0.057 token cost | 523ms | 14,103 | $0.005 / search |
| 4 | You | POST /v1/search extraction_mode=highlights | Official docs ↗ | 41.3 ± 3.1 | 24s | $0.097$0.025 search$0.072 token cost | 596ms | 20,333 | $0.005 / search |
| 5 | Exa fast | POST /search type=fast | Official docs ↗ | 66.3 ± 1.5 | 20s | $0.103$0.035 search$0.068 token cost | 626ms | 22,344 | $0.007 / search |
| 6 | You | POST /v1/search extraction_mode=highlights · knowledge=core | Official docs ↗ | 38.3 ± 3.2 | 21s | $0.096$0.025 search$0.071 token cost | 677ms | 20,786 | $0.005 / search |
| 7 | Nimble | POST /v2/search search_depth=standard · full_content=false · focus=general | Official docs ↗ | 42.3 ± 2.5 | 25s | $0.079$0.025 search$0.054 token cost | 878ms | 14,859 | $0.005 / search |
| 8 | Parallel fast | POST /v1/search mode=fast | Official docs ↗ | 66.7 ± 1.5 | 22s | $0.054$0.005 search$0.049 token cost | 953ms | 12,460 | $0.001 / search |
| 9 | Perplexity | POST /search search_context_size=low | Official docs ↗ | 77.3 ± 2.1 | 18s | $0.059$0.025 search$0.034 token cost | 957ms | 8,765 | $0.005 / search |
| 10 | Linkup fast | POST /v1/search depth=fast | Official docs ↗ | 43.3 ± 1.1 | 27s | $0.100$0.025 search$0.075 token cost | 1.39s | 24,057 | $0.005 / search |
| 11 | Tavily basic | POST /search search_depth=basic | Official docs ↗ | 51.0 ± 5.3 | 26s | $0.097$0.040 search$0.057 token cost | 1.72s | 16,299 | $0.008 / search |
| 12 | TinyFish | GET api.search.tinyfish.ai | Official docs ↗ | 59.3 ± 1.5 | 28s | $0.035$0.000 search$0.035 token cost | 2.15s | 7,469 | $0 / search |
| 13 | Firecrawl | POST /v2/search | Official docs ↗ | 70.3 ± 1.5 | 28s | $0.059$0.025 search$0.034 token cost | 2.87s | 7,456 | $0.005 / search |
| 14 | Nimble | POST /v2/search search_depth=lite · full_content=false · focus=general | Official docs ↗ | 46.7 ± 3.8 | 33s | $0.043$0.005 search$0.037 token cost | 3.15s | 8,713 | $0.0011 / search |
Developer search + scrape: latency, accuracy and cost
| Rank | Vendor | Endpoint & configuration | Official docs | Task completion | Median task time | Median task cost | Avg search time | Median task tokens | API list price |
|---|---|---|---|---|---|---|---|---|---|
| 1 | You | POST /v1/search extraction_mode=highlightsPOST /v1/contents | Official docs ↗ | 55.0 ± 1.7 | 27s | $0.119$0.027 search+scrape$0.092 token cost | 638ms | 42,806 | $0.005 / search$0.001 / scrape |
| 2 | You | POST /v1/search extraction_mode=highlights · knowledge=corePOST /v1/contents | Official docs ↗ | 54.0 ± 2.0 | 28s | $0.117$0.026 search+scrape$0.091 token cost | 678ms | 37,453 | $0.005 / search$0.001 / scrape |
| 3 | Nimble | POST /v2/search search_depth=standard · full_content=false · focus=generalPOST /v2/extract formats=[markdown] | Official docs ↗ | 45.0 ± 1.0 | 44s | $0.078$0.078 token cost | 905ms | 28,586 | $0.005 / search |
| 4 | Perplexity | POST /search search_context_size=highPOST /search search_context_size=high | Official docs ↗ | 77.7 ± 1.5 | 22s | $0.083$0.025 search+scrape$0.058 token cost | 991ms | 20,062 | $0.005 / search$0.005 / scrape |
| 5 | Exa auto | POST /search type=autoPOST /contents | Official docs ↗ | 81.7 ± 1.1 | 23s | $0.107$0.032 search+scrape$0.074 token cost | 1.19s | 27,433 | $0.007 / search$0.001 / scrape |
| 6 | TinyFish | GET api.search.tinyfish.aiGET api.fetch.tinyfish.ai format=markdown | Official docs ↗ | 79.0 ± 2.0 | 24s | $0.050$0.000 search+scrape$0.050 token cost | 1.32s | 12,844 | $0 / search$0 / scrape |
| 7 | Tavily basic | POST /search search_depth=basicPOST /extract extract_depth=basic | Official docs ↗ | 59.0 ± 1.7 | 32s | $0.113$0.043 search+scrape$0.070 token cost | 1.50s | 27,405 | $0.008 / search$0.0016 / scrape |
| 8 | Parallel basic | POST /v1/search mode=basicPOST /v1/extract | Official docs ↗ | 76.0 ± 0.0 | 30s | $0.107$0.023 search+scrape$0.084 token cost | 1.59s | 32,809 | $0.005 / search$0.001 / scrape |
| 9 | Nimble | POST /v2/search search_depth=lite · full_content=false · focus=generalPOST /v2/extract formats=[markdown] | Official docs ↗ | 60.3 ± 2.5 | 40s | $0.054$0.054 token cost | 1.93s | 16,828 | $0.0011 / search |
| 10 | Linkup standard | POST /v1/search depth=standardPOST /v1/fetch mode=standard | Official docs ↗ | 48.3 ± 4.9 | 48s | $0.147$0.038 search+scrape$0.108 token cost | 2.00s | 57,791 | $0.005 / search$0.005 / scrape |
| 11 | Firecrawl | POST /v2/searchPOST /v2/scrape | Official docs ↗ | 76.0 ± 1.0 | 34s | $0.082$0.028 search+scrape$0.055 token cost | 2.81s | 17,379 | $0.005 / search$0.0025 / scrape |
| 12 | Parallel advanced | POST /v1/search mode=advancedPOST /v1/extract | Official docs ↗ | 77.0 ± 1.0 | 33s | $0.096$0.019 search+scrape$0.077 token cost | 3.11s | 27,092 | $0.005 / search$0.001 / scrape |
| 13 | Tavily advanced | POST /search search_depth=advancedPOST /extract extract_depth=advanced | Official docs ↗ | 60.0 ± 2.0 | 46s | $0.153$0.085 search+scrape$0.069 token cost | 3.41s | 26,269 | $0.016 / search$0.0032 / scrape |
| 14 | Exa deep | POST /search type=deepPOST /contents | Official docs ↗ | 83.0 ± 1.0 | 37s | $0.127$0.057 search+scrape$0.070 token cost | 3.97s | 23,660 | $0.012 / search$0.001 / scrape |
Supporting Multiple Search Results
45 company-discovery questions with an agent making multiple searches. These tables keep the existing time-per-task-quality comparison: median agent time divided by F1 as a fraction. It measures the time needed relative to task quality.
Multiple searches: time per task quality
| Provider | Endpoint & configuration | Official docs | Time / task quality | Task quality (F1) | Median time |
|---|---|---|---|---|---|
| Exa instant | POST /search type=instantexa-instant | Official docs ↗ | 115.0 s | 43.3% | 49.8 s |
| You | POST /v1/searchextraction_mode=highlights | Official docs ↗ | 124.7 s | 38.6% | 48.2 s |
| You | POST /v1/searchextraction_mode=highlights · knowledge=core | Official docs ↗ | 125.0 s | 38.1% | 47.6 s |
| Perplexity | POST /searchsearch_context_size=low | Official docs ↗ | 129.4 s | 37.8% | 48.9 s |
| Parallel turbo | POST /v1/search mode=turboparallel-turbo | Official docs ↗ | 133.6 s | 34.7% | 46.4 s |
| Parallel fast | POST /v1/search mode=fastparallel-fast | Official docs ↗ | 141.8 s | 38.0% | 53.9 s |
| Parallel basic | POST /v1/search mode=basicparallel-basic | Official docs ↗ | 145.4 s | 46.5% | 67.6 s |
| Brave Search | GET /res/v1/web/searchbrave | Official docs ↗ | 155.5 s | 28.0% | 43.5 s |
| Linkup fast | POST /v1/search depth=fastlinkup-fast | Official docs ↗ | 156.0 s | 41.1% | 64.2 s |
| Nimble | POST /v2/searchsearch_depth=standard · full_content=false · focus=general | Official docs ↗ | 172.4 s | 30.7% | 53.0 s |
| Tavily basic | POST /search search_depth=basictavily-basic | Official docs ↗ | 172.7 s | 36.5% | 63.0 s |
| Linkup standard | POST /v1/search depth=standardlinkup-standard | Official docs ↗ | 178.2 s | 40.6% | 72.3 s |
| Parallel advanced | POST /v1/search mode=advancedparallel-advanced | Official docs ↗ | 188.9 s | 44.2% | 83.4 s |
| Exa deep | POST /search type=deepexa-deep | Official docs ↗ | 197.2 s | 45.4% | 89.5 s |
| Tavily advanced | POST /search search_depth=advancedtavily-advanced | Official docs ↗ | 224.4 s | 41.1% | 92.2 s |
| TinyFish | GET api.search.tinyfish.aitinyfish | Official docs ↗ | 241.0 s | 26.6% | 64.2 s |
| Firecrawl | POST /v2/searchfirecrawl | Official docs ↗ | 246.6 s | 30.4% | 75.0 s |
| Nimble | POST /v2/searchsearch_depth=lite · full_content=false · focus=general | Official docs ↗ | 278.4 s | 24.1% | 67.2 s |
| Seltz | POST /v1/search scope=companiesseltz-companies | Official docs ↗ | 379.7 s | 14.5% | 55.2 s |
| SERP (RapidAPI) | GET google-search74.p.rapidapi.comserp | Official docs ↗ | 8480.0 s | 0.4% | 31.4 s |
Time / task quality is median agent time divided by F1 (task completion quality). Ranked by time / task quality so a fast incomplete set costs more.
Multiple searches + scrape: time per task quality
| Provider | Endpoint & configuration | Official docs | Time / task quality | Task quality (F1) | Median time |
|---|---|---|---|---|---|
| Perplexity | POST /searchsearch_context_size=high | Official docs ↗ | 115.7 s | 46.6% | 53.9 s |
| Exa instant | POST /search type=instantexa-instant | Official docs ↗ | 117.0 s | 44.9% | 52.5 s |
| You | POST /v1/searchextraction_mode=highlights | Official docs ↗ | 118.4 s | 38.5% | 45.6 s |
| You | POST /v1/searchextraction_mode=highlights · knowledge=core | Official docs ↗ | 126.0 s | 38.5% | 48.5 s |
| Parallel turbo | POST /v1/search mode=turboparallel-turbo | Official docs ↗ | 133.4 s | 36.0% | 48.1 s |
| Parallel fast | POST /v1/search mode=fastparallel-fast | Official docs ↗ | 140.3 s | 39.3% | 55.1 s |
| Brave Search | GET /res/v1/web/searchbrave | Official docs ↗ | 153.4 s | 29.4% | 45.1 s |
| Parallel basic | POST /v1/search mode=basicparallel-basic | Official docs ↗ | 162.2 s | 42.3% | 68.6 s |
| Tavily basic | POST /search search_depth=basictavily-basic | Official docs ↗ | 162.4 s | 38.1% | 62.0 s |
| Linkup fast | POST /v1/search depth=fastlinkup-fast | Official docs ↗ | 170.5 s | 39.9% | 68.0 s |
| Nimble | POST /v2/searchsearch_depth=standard · full_content=false · focus=general · extract formats=[markdown] | Official docs ↗ | 189.1 s | 31.3% | 59.2 s |
| Parallel advanced | POST /v1/search mode=advancedparallel-advanced | Official docs ↗ | 191.6 s | 42.2% | 80.9 s |
| Linkup standard | POST /v1/search depth=standardlinkup-standard | Official docs ↗ | 192.8 s | 42.0% | 81.0 s |
| Exa deep | POST /search type=deepexa-deep | Official docs ↗ | 198.9 s | 48.2% | 95.8 s |
| TinyFish | GET api.search.tinyfish.aitinyfish | Official docs ↗ | 209.6 s | 30.2% | 63.4 s |
| Tavily advanced | POST /search search_depth=advancedtavily-advanced | Official docs ↗ | 225.9 s | 41.0% | 92.6 s |
| Firecrawl | POST /v2/searchfirecrawl | Official docs ↗ | 247.2 s | 33.2% | 82.1 s |
| Nimble | POST /v2/searchsearch_depth=lite · full_content=false · focus=general · extract formats=[markdown] | Official docs ↗ | 280.3 s | 25.8% | 72.4 s |
| Seltz | POST /v1/search scope=companiesseltz-companies | Official docs ↗ | 369.4 s | 16.3% | 60.1 s |
| SERP (RapidAPI) | GET google-search74.p.rapidapi.comserp | Official docs ↗ | — | 0.0% | 33.4 s |
Time / task quality is median agent time divided by F1 (task completion quality). Ranked by time / task quality so a fast incomplete set costs more.
Fast Web Search API FAQ
What is the fastest web search API for AI agents?
Parallel turbo is fastest by mean search response time on this 300-question benchmark at 348ms. It scores 71.3% accuracy and has a listed search price of $1.00 per 1,000 queries. The primary table compares the same search questions across all tested modes.
Which fast search API has high accuracy?
Exa instant is the fastest tested mode reaching at least 95% accuracy: 398ms average response time and 97.7% accuracy. Exa fast leads overall search accuracy at 99.3%, with 652ms average latency. The 95% threshold is a shortlist filter; the main ranking remains sorted by average latency.
Is Parallel turbo faster than Exa instant?
Parallel turbo: 348ms average search time, 71.3% accuracy and $1.00 per 1,000 queries. Exa instant: 398ms average search time, 97.7% accuracy and $7.00 per 1,000 queries. These are means from the same questions and harness. Compare this measured speed and accuracy tradeoff with the latency target for your own queries.
How do Tavily, Brave and Firecrawl compare on search speed?
Firecrawl: 510ms average search time, 95.3% accuracy and $5.00 per 1,000 queries. Brave Search llm-context: 601ms average search time, 94.0% accuracy and $5.00 per 1,000 queries. Brave Search web: 630ms average search time, 93.3% accuracy and $5.00 per 1,000 queries. Tavily basic: 1.88s average search time, 87.7% accuracy and $8.00 per 1,000 queries. Tavily advanced: 4.29s average search time, 93.0% accuracy and $16.00 per 1,000 queries. Firecrawl’s listed rate is for Standard-plan top-ups and requires a subscription; Tavily uses its pay-as-you-go credit rate.
What is the difference between mean latency, p50 and p95?
Mean latency is the average response time across measured queries. Median, or p50, is the middle response time. P95 describes the slowest 5% boundary and helps size a production timeout. This page ranks by measured mean and shows median separately; collect p95 and p99 from your application’s query and concurrency profile when choosing production settings.
Why do vendor latency claims differ from this benchmark?
Latency depends on the query set, deployment region, result count, returned content, cache state, concurrency and search mode. Compare the same statistic under matching conditions: a vendor p50 and a benchmark mean summarize different things. This table uses one benchmark’s measurements, with endpoint settings and official docs linked for each mode.
What does this search benchmark measure?
Each API receives the same 300 company-news questions, with one search per question and up to 10 results. Search time measures the API request; accuracy is the share of questions answered correctly from its returned evidence, using the same answer model and judge. The code and public dataset sample are available, with a held-out scoring set.
How much do the fastest search APIs cost per 1,000 queries?
Parallel turbo costs $1.00 per 1,000 queries at its listed rate. Exa instant costs $7.00 per 1,000 queries at its listed rate. Exa fast costs $7.00 per 1,000 queries at its listed rate. The main table shows billing conditions, official pricing and free limits. Additional results, summaries, scrape calls and LLM tokens affect the application budget.
How do I choose a low-latency web search API for RAG or LLM apps?
Choose a search mode that meets your latency budget and returns evidence your model can use. Check accuracy and price alongside mean and median response time. Then test representative queries from your deployment region at expected concurrency, including slow requests, retries and rate limits. For agents making several searches, use the supporting task-time tables below.
How is search response time different from full agent-task time?
Search response time covers one API request. Full agent-task time also includes model generation, repeated searches and optional scrape calls. The supporting developer benchmark uses 100 questions and reports average search time plus median task time. The multiple-search benchmark uses 45 company-discovery questions and ranks by median agent time divided by F1 as a fraction.