100tasks · hard retrieval · search & fetch and search-only
Most token-efficient web search API for coding agents
Ranked on the hard retrieval task of the web search benchmark. A pass requires a grounded source URL from that run's search or fetch. Rankings do not transfer from factual lookup or multi-hop.
Firecrawl is most token-efficient on search & fetch at 17,379 tokens. Firecrawl is most token-efficient on search-only at 7,456 tokens.
Most token-efficient on this board is median task tokens, not search-API list price. Search & fetch and search-only are separate rankings.
Most accurate · Fastest · Most token efficient · Full benchmark
[01]search & fetch
Ranking, sorted by median task tokens
| Rank | Vendor | Endpoint & configuration | Task completion | Median task time | Avg search time | Median task tokens |
|---|---|---|---|---|---|---|
| 1 | Firecrawl | POST /v2/searchPOST /v2/scrape | 76.0 ± 1.0 | 34s | 2.81s | 17,379 |
| 2 | Exa deep | POST /search type=deepPOST /contents | 83.0 ± 1.0 | 37s | 3.97s | 23,660 |
| 3 | Tavily advanced | POST /search search_depth=advancedPOST /extract extract_depth=advanced | 60.0 ± 2.0 | 46s | 3.41s | 26,269 |
| 4 | Tavily basic | POST /search search_depth=basicPOST /extract extract_depth=basic | 59.0 ± 1.7 | 32s | 1.50s | 27,405 |
| 5 | Exa auto | POST /search type=autoPOST /contents | 81.7 ± 1.1 | 23s | 1.19s | 27,433 |
| 6 | Parallel advanced | POST /v1/search mode=advancedPOST /v1/extract | 79.3 ± 2.1 | 39s | 3.08s | 29,925 |
| 7 | Parallel basic | POST /v1/search mode=basicPOST /v1/extract | 73.7 ± 2.3 | 37s | 1.70s | 41,475 |
| 8 | Linkup standard | POST /v1/search depth=standardPOST /v1/fetch mode=standard | 48.3 ± 4.9 | 48s | 2.00s | 57,791 |
[02] search only
Search-only, same sort
| Rank | Vendor | Endpoint & configuration | Task completion | Median task time | Avg search time | Median task tokens |
|---|---|---|---|---|---|---|
| 1 | Firecrawl | POST /v2/search | 70.3 ± 1.5 | 28s | 2.87s | 7,456 |
| 2 | Parallel fast | POST /v1/search mode=fast | 63.0 ± 3.6 | 22s | 954ms | 14,747 |
| 3 | Parallel turbo | POST /v1/search mode=turbo | 55.3 ± 1.5 | 21s | 454ms | 17,820 |
| 4 | Brave Search | POST /res/v1/llm/context | 43.0 ± 2.0 | 23s | 547ms | 21,000 |
| 5 | Exa fast | POST /search type=fast | 66.3 ± 1.5 | 20s | 626ms | 22,344 |
| 6 | Exa instant | POST /search type=instant | 61.3 ± 2.9 | 21s | 447ms | 22,423 |
| 7 | Tavily fast | POST /search search_depth=fast | 47.3 ± 2.5 | 22s | 282ms | 23,922 |
| 8 | Linkup fast | POST /v1/search depth=fast | 43.3 ± 1.1 | 27s | 1.39s | 24,057 |
[03] faq
Common questions
What is the most token-efficient web search API for coding agents?
Firecrawl is most token-efficient on search & fetch at 17,379 tokens. Firecrawl is most token-efficient on search-only at 7,456 tokens. That is the LLM bill for the coding loop, not search-API list price.