Best Web Search APIs for AI Agents in 2026: Independent Benchmarks and Comparison
Parallel basic at 46.5% F1 leads these Independent 2026 benchmarks of the best web search APIs for AI agents across 12 providers, including Exa, Parallel, Tavily, Brave and Firecrawl. Sub-benchmark, single-query answer accuracy: Exa fast at 99.3%. Sub-benchmark, RAG tasks: Perplexity at 77.3% completion. Fastest single search: Parallel turbo at 348ms mean response. Lowest paid list price: Parallel at $1 per 1,000 queries. Three separate web search tasks.
- Headline benchmark, agent search: Parallel basic at 46.5% F1. Several searches, answer quality. Agent search table.
- Sub-benchmark, single-query answer accuracy: Exa fast at 99.3%. One search, one answer. Single-query table.
- Sub-benchmark, RAG tasks: Perplexity at 77.3% completion. RAG table.
- Fastest single search: Parallel turbo at 348ms mean response, with 71.3% single-query accuracy. This is one API call, not a full agent task.
- Lowest paid list price: Parallel at $1 per 1,000 queries. At 100,000 searches that is $100. Paid pricing table.
Independent benchmarks, not one accuracy number
Vendor benchmarks pick the questions, the metric, and the rival they compare against. A 99% score on one search and a mid-40s F1 on an agent that searches several times can both be real, and they are not the same claim. This page runs three tests. One is the headline for AI agents. The other two are sub-benchmarks with their own winners.
- Agent search, the headline benchmark. The agent may search more than once and must return the right set of companies. The score is F1: precision and recall together. Use this when you are choosing a search API for an AI agent.
- Single-query answer accuracy, a sub-benchmark. One question, one search, up to 10 results. A fixed model extracts the answer. The score is the share of answers that match the label. Use this when the agent looks up one fact per call.
- RAG tasks, a sub-benchmark. Held-out tickets answered from retrieved passages. The score is grounded task completion from search results. Use this when the agent retrieves context and answers from it.
- Fasteston this page means mean response time of one search call on the single-query test, reported with that call's accuracy. It does not mean the shortest full agent task.
- Cheapest on this page means the lowest paid list price per 1,000 queries. Free-tier $0 rates are left out. At 100,000 searches, multiply the per-1,000 price by 100.
Best search API for AI agents in 2026: agent search
F1 on 45 questions that need several searches. Same agent, same questions, search results only. F1 balances finding the right answers with finding the complete set. This is the ranking to cite for an AI agent. It is not an accuracy percentage, and it is not comparable to the sub-benchmarks below.
Full multi-search benchmark → · Search and page-scrape variant →
| Provider | Endpoint & configuration | Official docs | F1 | Precision | Recall | Exact set | Median time | Mean turns | Median task cost | API list price |
|---|---|---|---|---|---|---|---|---|---|---|
| Parallel basic | POST /v1/search mode=basicparallel-basic | Parallel basic official docs | 46.5 ± 1.9 | 88.7 ± 0.9 | 34.4 ± 2.0 | 3.7 ± 2.6 | 67.6 s | 7.4 | $1.120$0.070 search$1.050 token cost | $0.005 / search |
| Exa deep | POST /search type=deepexa-deep | Exa deep official docs | 45.4 ± 2.0 | 83.2 ± 3.0 | 33.7 ± 1.4 | 1.5 ± 1.3 | 89.5 s | 7.4 | $0.717$0.156 search$0.556 token cost | $0.012 / search |
| Parallel advanced | POST /v1/search mode=advancedparallel-advanced | Parallel advanced official docs | 44.2 ± 1.4 | 87.6 ± 3.4 | 32.0 ± 1.6 | 2.2 ± 0.0 | 83.4 s | 7.3 | $0.625$0.070 search$0.565 token cost | $0.005 / search |
| Exa instant | POST /search type=instantexa-instant | Exa instant official docs | 43.3 ± 1.0 | 82.6 ± 3.8 | 32.2 ± 0.5 | 3.0 ± 1.3 | 49.8 s | 7.3 | $0.652$0.091 search$0.559 token cost | $0.007 / search |
| Linkup fast | POST /v1/search depth=fastlinkup-fast | Linkup fast official docs | 41.1 ± 1.7 | 82.8 ± 1.0 | 30.3 ± 1.5 | 0.7 ± 1.3 | 64.2 s | 7.4 | $0.923$0.070 search$0.853 token cost | $0.005 / search |
| Tavily advanced | POST /search search_depth=advancedtavily-advanced | Tavily advanced official docs | 41.1 ± 2.3 | 83.7 ± 4.2 | 29.8 ± 1.7 | 2.2 ± 0.0 | 92.2 s | 7.4 | $1.029$0.224 search$0.808 token cost | $0.016 / search |
| Linkup standard | POST /v1/search depth=standardlinkup-standard | Linkup standard official docs | 40.6 ± 0.9 | 84.0 ± 7.9 | 29.7 ± 1.4 | 1.5 ± 2.6 | 72.3 s | 7.3 | $0.937$0.070 search$0.867 token cost | $0.005 / search |
| You | POST /v1/searchextraction_mode=highlights | You official docs | 38.6 ± 0.8 | 79.3 ± 4.9 | 28.3 ± 0.9 | 4.4 ± 0.0 | 48.2 s | 7.5 | $0.971$0.070 search$0.901 token cost | $0.005 / search |
| You | POST /v1/searchextraction_mode=highlights · knowledge=core | You official docs | 38.1 ± 0.6 | 80.9 ± 1.6 | 27.4 ± 0.9 | 3.0 ± 1.3 | 47.6 s | 7.5 | $0.895$0.070 search$0.827 token cost | $0.005 / search |
| Parallel fast | POST /v1/search mode=fastparallel-fast | Parallel fast official docs | 38.0 ± 2.0 | 79.9 ± 4.6 | 27.5 ± 1.4 | 1.5 ± 1.3 | 53.9 s | 7.6 | $0.460$0.014 search$0.447 token cost | $0.001 / search |
| Perplexity | POST /searchsearch_context_size=low | Perplexity official docs | 37.8 ± 2.1 | 79.3 ± 5.7 | 26.8 ± 1.4 | 2.2 ± 2.2 | 48.9 s | 7.7 | $0.334$0.070 search$0.264 token cost | $0.005 / search |
| Tavily basic | POST /search search_depth=basictavily-basic | Tavily basic official docs | 36.5 ± 3.2 | 82.5 ± 1.9 | 25.7 ± 2.7 | 1.5 ± 1.3 | 63.0 s | 7.7 | $0.615$0.112 search$0.504 token cost | $0.008 / search |
| Parallel turbo | POST /v1/search mode=turboparallel-turbo | Parallel turbo official docs | 34.7 ± 2.4 | 80.0 ± 5.2 | 24.8 ± 2.4 | 1.5 ± 1.3 | 46.4 s | 7.6 | $0.419$0.014 search$0.405 token cost | $0.001 / search |
| Nimble | POST /v2/searchsearch_depth=standard · full_content=false · focus=general | Nimble official docs | 30.7 ± 1.4 | 70.3 ± 3.3 | 21.3 ± 1.1 | 0.7 ± 1.3 | 53.0 s | 7.7 | $0.467$0.070 search$0.398 token cost | $0.005 / search |
| Firecrawl | POST /v2/searchfirecrawl | Firecrawl official docs | 30.4 ± 1.1 | 77.3 ± 4.4 | 20.7 ± 0.7 | 2.2 ± 0.0 | 75.0 s | 7.9 | $0.282$0.070 search$0.212 token cost | $0.005 / search |
| Brave Search | GET /res/v1/web/searchbrave | Brave Search official docs | 28.0 ± 1.7 | 66.9 ± 8.5 | 19.3 ± 0.5 | 0.7 ± 1.3 | 43.5 s | 8.0 | $0.268$0.070 search$0.200 token cost | $0.005 / search |
| TinyFish | GET api.search.tinyfish.aitinyfish | TinyFish official docs | 26.6 ± 1.3 | 64.7 ± 1.5 | 17.9 ± 0.8 | 1.5 ± 1.3 | 64.2 s | 7.9 | $0.212$0.000 search$0.212 token cost | $0 / search |
| Nimble | POST /v2/searchsearch_depth=lite · full_content=false · focus=general | Nimble official docs | 24.1 ± 1.4 | 65.2 ± 2.3 | 16.1 ± 1.3 | 0.7 ± 1.3 | 67.2 s | 7.9 | $0.222$0.015 search$0.207 token cost | $0.0011 / search |
| Seltz | POST /v1/search scope=companiesseltz-companies | Seltz official docs | 14.5 ± 0.9 | 40.0 ± 3.1 | 9.4 ± 0.7 | 0.0 ± 0.0 | 55.2 s | 7.3 | $1.751$0.070 search$1.681 token cost | $0.005 / search |
| SERP (RapidAPI) | GET google-search74.p.rapidapi.comserp | SERP (RapidAPI) official docs | 0.4 ± 0.6 | 0.7 ± 1.3 | 0.3 ± 0.4 | 0.0 ± 0.0 | 31.4 s | 7.8 | $0.103$0.036 search$0.065 token cost | $0.003 / search |
F1, precision, recall, and exact-set accuracy are percentages reported as mean ± sample SD across three independent runs; each run aggregates all 45 questions. SD is measured in percentage points. Median time is the median end-to-end time across all runs for each vendor. Median task cost is the median of LLM $ plus search/scrape API $ per agent run. API list price is the PAYG unit rate of the search (and scrape) endpoint the harness calls.
Sub-benchmark: single-query answer accuracy
Extracted-answer accuracy on 300 company-news questions. One query, up to 10 results, same extraction model and judge. Mean response time in this table is the fastest-single-search measurement. Do not quote this percentage as the winner of the agent-search test.
| Rank | Vendor | Endpoint & configuration | Official docs | Accuracy | AR@1 | AR@5 | $ / 1k correct | Latency | Snippet tokens | Total $ | Cost per 1,000 queries |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Exaweb search | POST /searchtype=fast | Exa fast official docs | 99.3% | 95.0% | 99.3% | $7.05 | 652ms | 1,987 | $2.10 | $7 / 1ktype=fast · up to 10 results |
| 2 | Exaweb search | POST /searchtype=instant | Exa instant official docs | 97.7% | 80.0% | 97.3% | $7.17 | 398ms | 2,128 | $2.10 | $7 / 1ktype=instant · up to 10 results |
| 3 | Perplexityweb search | POST /searchsearch_context_size=low | Perplexity low official docs | 97.3% | 91.7% | 98.3% | $5.14 | 1.38s | 476 | $1.50 | $5 / 1kSearch API · POST /search |
| 4 | Linkupweb search | POST /v1/searchdepth=fast · outputType=searchResults | Linkup fast official docs | 96.7% | 79.0% | 94.7% | $5.17 | 1.57s | 3,022 | $1.50 | $5 / 1kdepth=fast · searchResults |
| 5 | SERPgoogle search | GET google-search74.p.rapidapi.comlimit=10 | SERP (RapidAPI) official docs | 96.0% | 78.0% | 95.0% | $3.13 | 751ms | 497 | $0.90 | $3 / 1kPro overage $0.003/request |
| 6 | Firecrawlweb search | POST /v2/search | Firecrawl official docs | 95.3% | 77.7% | 96.7% | $5.24 | 510ms | 678 | $1.50 | $5 / 1k2 credits / 10 results |
| 7 | Brave Searchweb search | POST /res/v1/llm/contextcount=10 | Brave LLM context official docs | 94.0% | 81.0% | 94.7% | $5.32 | 601ms | 2,064 | $1.50 | $5 / 1kSearch plan · LLM Context |
| 8 | Parallelweb search | POST /v1/searchmode=basic | Parallel basic official docs | 93.3% | 55.0% | 92.0% | $5.36 | 1.68s | 2,330 | $1.50 | $5 / 1kmode=basic · 10 results |
| 9 | Brave Searchweb search | GET /res/v1/web/searchcount=10 · result_filter=web | Brave Search official docs | 93.3% | 79.3% | 91.7% | $5.36 | 630ms | 817 | $1.50 | $5 / 1k |
| 10 | Nimbleweb search | POST /v2/searchsearch_depth=standard · full_content=false · focus=general | Nimble official docs | 93.0% | 86.3% | 94.3% | $5.38 | 861ms | 2,730 | $1.50 | $5 / 1ksearch_depth=standard · full_content=false |
| 11 | Tavilyweb search | POST /searchsearch_depth=advanced | Tavily advanced official docs | 93.0% | 65.0% | 92.7% | $17.20 | 4.29s | 2,210 | $4.80 | $16 / 1k2 credits · $0.008 PAYG |
| 12 | TinyFishweb search | GET api.search.tinyfish.aifree · 30 req/min cap | TinyFish official docs | 92.0% | 74.3% | 90.7% | Free30 req/min cap | 2.62s | 441 | Free | $0free · 30 req/min, 500/hour |
| 13 | Linkupweb search | POST /v1/searchdepth=standard · outputType=searchResults | Linkup standard official docs | 92.0% | 67.3% | 90.3% | $5.43 | 2.55s | 2,983 | $1.50 | $5 / 1kdepth=standard · searchResults |
| 14 | Youweb search | POST /v1/searchextraction_mode=highlights · knowledge=core | You official docs | 92.0% | 67.3% | 89.7% | $5.43 | 889ms | 2,862 | $1.50 | $5 / 1khighlights · knowledge=core · same Web Search price assumption |
| 15 | Youweb search | POST /v1/searchextraction_mode=highlights | You official docs | 90.7% | 72.0% | 89.7% | $5.51 | 628ms | 2,837 | $1.50 | $5 / 1khighlights included in Web Search |
| 16 | Tavilyweb search | POST /searchsearch_depth=basic | Tavily basic official docs | 87.7% | 82.0% | 87.7% | $9.13 | 1.88s | 1,639 | $2.40 | $8 / 1k1 credit · $0.008 PAYG |
| 17 | Parallelweb search | POST /v1/searchmode=fast | Parallel fast official docs | 86.0% | 44.3% | 79.0% | $1.16 | 942ms | 1,839 | $0.30 | $1 / 1kmode=fast · 10 results |
| 18 | Nimbleweb search | POST /v2/searchsearch_depth=lite · full_content=false · focus=general | Nimble official docs | 75.3% | 64.0% | 75.3% | $1.46 | 3.15s | 413 | $0.33 | $1.1 / 1ksearch_depth=lite · full_content=false |
| 19 | Parallelweb search | POST /v1/searchmode=turbo | Parallel turbo official docs | 71.3% | 45.3% | 66.0% | $1.40 | 348ms | 1,853 | $0.30 | $1 / 1kmode=turbo · 10 results |
$ / 1k correct is $ / 1k queries divided by extracted-answer accuracy. Ranked by accuracy, highest first. Cost per 1,000 queries is the published PAYG list price, linked to the vendor pricing page. Not promotional packs or volume discounts. TinyFish Search is free with a 30 requests/min cap — $0 is not unlimited throughput.
Sub-benchmark: RAG tasks
Grounded task completion on 100 held-out tickets, search results only. The winner here is the best API for this RAG job. It is not the best API for agent search, and it is not the single-query accuracy leader.
Full documentation-search benchmark → · Developer comparison →
| Rank | Vendor | Endpoint & configuration | Official docs | Task completion | Median task time | Median task cost | Avg search time | Median task tokens | API list price |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Perplexity | POST /search search_context_size=low | Perplexity official docs | 77.3 ± 2.1 | 18s | $0.059$0.025 search$0.034 token cost | 957ms | 8,765 | $0.005 / search |
| 2 | Firecrawl | POST /v2/search | Firecrawl official docs | 70.3 ± 1.5 | 28s | $0.059$0.025 search$0.034 token cost | 2.87s | 7,456 | $0.005 / search |
| 3 | Parallel fast | POST /v1/search mode=fast | Parallel fast official docs | 66.7 ± 1.5 | 22s | $0.054$0.005 search$0.049 token cost | 953ms | 12,460 | $0.001 / search |
| 4 | Exa fast | POST /search type=fast | Exa fast official docs | 66.3 ± 1.5 | 20s | $0.103$0.035 search$0.068 token cost | 626ms | 22,344 | $0.007 / search |
| 5 | Parallel turbo | POST /v1/search mode=turbo | Parallel turbo official docs | 64.7 ± 2.1 | 19s | $0.059$0.005 search$0.054 token cost | 333ms | 14,130 | $0.001 / search |
| 6 | Exa instant | POST /search type=instant | Exa instant official docs | 61.3 ± 2.9 | 21s | $0.105$0.035 search$0.070 token cost | 447ms | 22,423 | $0.007 / search |
| 7 | TinyFish | GET api.search.tinyfish.ai | TinyFish official docs | 59.3 ± 1.5 | 28s | $0.035$0.000 search$0.035 token cost | 2.15s | 7,469 | $0 / search |
| 8 | Tavily basic | POST /search search_depth=basic | Tavily basic official docs | 51.0 ± 5.3 | 26s | $0.097$0.040 search$0.057 token cost | 1.72s | 16,299 | $0.008 / search |
| 9 | Nimble | POST /v2/search search_depth=lite · full_content=false · focus=general | Nimble official docs | 46.7 ± 3.8 | 33s | $0.043$0.005 search$0.037 token cost | 3.15s | 8,713 | $0.0011 / search |
| 10 | Linkup fast | POST /v1/search depth=fast | Linkup 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 | Nimble | POST /v2/search search_depth=standard · full_content=false · focus=general | Nimble official docs | 42.3 ± 2.5 | 25s | $0.079$0.025 search$0.054 token cost | 878ms | 14,859 | $0.005 / search |
| 12 | You | POST /v1/search extraction_mode=highlights | You official docs | 41.3 ± 3.1 | 24s | $0.097$0.025 search$0.072 token cost | 596ms | 20,333 | $0.005 / search |
| 13 | You | POST /v1/search extraction_mode=highlights · knowledge=core | You official docs | 38.3 ± 3.2 | 21s | $0.096$0.025 search$0.071 token cost | 677ms | 20,786 | $0.005 / search |
| 14 | Brave | POST /res/v1/llm/context | Brave official docs | 38.0 ± 2.6 | 20s | $0.082$0.025 search$0.057 token cost | 523ms | 14,103 | $0.005 / search |
Task completion is mean ± SD of 3 runs; n = 100 tasks.
Paid web search API price per 1,000 queries
Lowest paid list price: Parallel at $1 per 1,000 queries, on Parallel fast (86.0% single-query accuracy) and Parallel turbo (71.3% single-query accuracy). At 100,000 searches that list price is $100.
Recorded list prices, checked on 1 September 2026. A $0 free-tier rate is omitted: a request cap is not a price at production volume. The rate is per search. An agent that runs several searches pays it on every search. Plans, credits, and extra results can change the bill. LLM tokens are separate.
| API configuration | Price / 1,000 queries | 100,000 searches | Single-query accuracy | Mean response | Official pricing |
|---|---|---|---|---|---|
| Parallel fast | $1 | $100 | 86.0% | 942ms | Official pricing ↗ |
| Parallel turbo | $1 | $100 | 71.3% | 348ms | Official pricing ↗ |
| Nimble lite | $1.10 | $110 | 75.3% | 3.15s | Official pricing ↗ |
| SERP | $3 | $300 | 96.0% | 751ms | Official pricing ↗ |
| Perplexity | $5 | $500 | 97.3% | 1.38s | Official pricing ↗ |
| Linkup fast | $5 | $500 | 96.7% | 1.57s | Official pricing ↗ |
| Firecrawl | $5 | $500 | 95.3% | 510ms | Official pricing ↗ |
| Brave Search llm-context | $5 | $500 | 94.0% | 601ms | Official pricing ↗ |
| Parallel basic | $5 | $500 | 93.3% | 1.68s | Official pricing ↗ |
| Brave Search web | $5 | $500 | 93.3% | 630ms | Official pricing ↗ |
| Nimble standard | $5 | $500 | 93.0% | 861ms | Official pricing ↗ |
| Linkup standard | $5 | $500 | 92.0% | 2.55s | Official pricing ↗ |
| You | $5 | $500 | 92.0% | 889ms | Official pricing ↗ |
| You | $5 | $500 | 90.7% | 628ms | Official pricing ↗ |
| Exa fast | $7 | $700 | 99.3% | 652ms | Official pricing ↗ |
| Exa instant | $7 | $700 | 97.7% | 398ms | Official pricing ↗ |
| Tavily basic | $8 | $800 | 87.7% | 1.88s | Official pricing ↗ |
| Tavily advanced | $16 | $1,600 | 93.0% | 4.29s | Official pricing ↗ |
How to choose the best search API for AI agents in 2026
Pick the benchmark that matches the call your agent makes. Then check speed and paid price on that same job. A provider can lead one test and sit mid-pack on another.
- The agent searches several times and combines evidence. Read the agent-search F1 table. Compare task time and task cost in that table only. This is the headline result for a search API for AI agents.
- The agent needs one fact from one search. Read single-query answer accuracy, then mean response time and paid price per 1,000 queries.
- The agent retrieves passages and answers from them. Read the RAG tasks sub-benchmark. Ignore the agent-search winner unless the agent also does that job.
- Then check the endpoint. Official docs in each table show the request that was measured. Rerun a sample of your own questions before you commit.
Exa vs Tavily · Tavily vs Parallel · Brave Search vs Exa · Linkup vs Firecrawl · Parallel vs Exa · Linkup vs Tavily · Exa vs Perplexity · Brave Search vs Tavily · Exa vs Firecrawl · Brave Search vs Parallel · LLM agents · LLM apps · Full web search benchmark
What each independent benchmark measures
- Agent search. 45 company-discovery questions, each with several constraints. The same agent and tool budget run against every API. Search results only on this page. Score: F1. Open runner · Public dataset.
- Single-query answers. 300 company-news questions, one natural-language query, up to 10 results. Score: extracted-answer accuracy. Speed: mean response time of that search call. Open code · Public dataset.
- RAG tasks. 100 held-out tickets, search results only, three runs. Score: grounded task completion. Open code · Public dataset.
Best web search API for AI agents: FAQ
What is the best search API for AI agents in 2026?
For an AI agent that runs several searches and combines the evidence, start with the headline benchmark: Parallel basic at 46.5% F1 on 45 questions. That is a different test from single-query answer accuracy, where Exa fast leads at 99.3% on 300 questions. It is also different from RAG tasks, where Perplexity leads at 77.3% completion on 100 tickets. Use the test that matches the agent. Do not average these scores.
What is the best web search API?
There is no single best web search API across every job. This page keeps three independent benchmarks separate. Agent search, several searches: Parallel basic at 46.5% F1. Single-query answer accuracy: Exa fast at 99.3%. RAG tasks: Perplexity at 77.3% completion. Teams choosing a search API for an AI agent should read the agent-search result first, then check single-search response time and paid price per 1,000 queries.
What is the best web search API for agents?
Parallel basic leads the headline benchmark for agents, at 46.5% F1, when the agent issues several searches and the score is answer quality. Single-query accuracy and RAG tasks are sub-benchmarks on this page. They answer a different question and can name a different provider. Read the label on each result before comparing them.
What is the fastest web search API for AI agents?
Parallel turbo has the lowest mean response time for one search call, at 348ms, with 71.3% single-query answer accuracy. That figure is the API's response to one search. It is not the time for a full agent task, which includes model turns and repeated searches. Compare task time only inside the agent-search table.
What is the cheapest web search API for AI agents?
The lowest paid list price is $1 per 1,000 queries, on Parallel fast (86.0% single-query accuracy) and Parallel turbo (71.3% single-query accuracy). At 100,000 searches that rate is $100. Prices are recorded list rates checked on 1 September 2026. A $0 free-tier rate is excluded, because a request cap is not a production price. An agent that searches several times pays the per-search rate on every search. Confirm current plans on the provider's pricing page.
Why do web search API roundups name different winners?
They are usually scoring different jobs and then calling each one accuracy. A single-query answer score, a RAG task score, and an agent-search F1 score are not the same measurement. A fast response on one search is not the same as a fast agent task. This page gives each claim one methodology and labels the other two results as sub-benchmarks.
How do I choose a web search API for an AI agent?
Match the benchmark to the agent. If the agent plans several searches and has to assemble an answer, use agent-search F1, then task time and task cost in that same table. If the agent needs one fact from one search, use single-query answer accuracy, mean response time, and paid price per 1,000 queries. If the agent retrieves passages and answers from them, use the RAG tasks sub-benchmark. Then read the official docs for the endpoint you would actually call, and rerun a sample of your own questions.
Are these web search API benchmarks independent?
Yes. Rank is not paid placement. Inside each test the questions, model, and judge stay fixed and the search API changes. Agent search uses 45 company-discovery questions that need several searches. Single-query answers use 300 company-news questions, one query, and up to 10 results. RAG tasks use 100 held-out tickets. Scoring questions are held out. Code and a public dataset sample are linked on this page.