Average search time: Web Search for Coding Agents, search only
Parallel turbo searches fastest at 333 msOf the 14 web search API configurations tested in search-only mode on the Web Search for Coding Agents Benchmark, Parallel turbo had the fastest average search time: 333 ms, pooled across all three runs of the 100-ticket set.
A coding agent patches a small httpx client against the vendor's current official API documentation on 100 held-out enterprise SaaS tickets, using each vendor's web search API. A ticket passes only when the patch compiles and every ground-truth token is grounded in a URL the search returned. Only the search API changes between rows.
Average search time is the mean latency of each web_search tool call, pooled across all three runs. Lower is better.
The agent has web_search only: it reads titles, URLs and snippets, and cannot open a page.
| Rank | Vendor | Endpoint & configuration | Avg search time |
|---|---|---|---|
| 1 | Parallel turbo | POST /v1/search mode=turbo | 333ms |
| 2 | Exa instant | POST /search type=instant | 447ms |
| 3 | Brave | POST /res/v1/llm/context | 523ms |
| 4 | You | POST /v1/search extraction_mode=highlights | 596ms |
| 5 | Exa fast | POST /search type=fast | 626ms |
| 6 | You | POST /v1/search extraction_mode=highlights · knowledge=core | 677ms |
| 7 | Nimble | POST /v2/search search_depth=standard · full_content=false · focus=general | 878ms |
| 8 | Parallel fast | POST /v1/search mode=fast | 953ms |
| 9 | Perplexity | POST /search search_context_size=low | 957ms |
| 10 | Linkup fast | POST /v1/search depth=fast | 1.39s |
| 11 | Tavily basic | POST /search search_depth=basic | 1.72s |
| 12 | TinyFish | GET api.search.tinyfish.ai | 2.15s |
| 13 | Firecrawl | POST /v2/search | 2.87s |
| 14 | Nimble | POST /v2/search search_depth=lite · full_content=false · focus=general | 3.15s |
