What is the /research endpoint and when should you use it?
Linkup Technical Staff
Linkup offers two web search API endpoints for LLMs: sub-second /search and multi-step /research. Here is when to use each.
What /search does and what it costs in latency
/search is a single-pass web search API for LLM applications that returns results in under a second. It ranks #1 among sub-second web search APIs on Verified SimpleQA with a 92% F-score, measured on the open eval harness at github.com/LinkupPlatform/eval-simpleQA.
Use /search when:
- The query maps to a fact that exists on one or two pages (a company's funding round, a product spec, a current price).
- Latency sits in a user-facing path. A chatbot waiting three seconds loses the user.
- You are doing RAG retrieval and the LLM will do the reasoning downstream.
/search is the table-stakes case done right: low latency, high accuracy on direct questions. It is the default for agent tool calls and live chat.
What /research does and why it takes longer
/research runs multiple search steps, reads across sources, and synthesises an answer to questions that no single page answers. It ranks #1 across the board on SealQA-0 with 61% accuracy, a benchmark built specifically for hard, multi-hop questions where naive search fails.
Use /research when:
- The answer requires combining facts from several sources ("compare the AI regulation timelines across the EU, US, and UK").
- The question is ambiguous and needs the system to decompose it before searching.
- Accuracy on hard questions matters more than response time, for example in a report-generation or due-diligence workflow.
/research trades latency for depth. It is not for the user-facing path. It is for the asynchronous job that produces a researched answer a human or agent will act on.
How to choose: a decision table
The choice comes down to two variables: how many sources the answer needs, and where the call sits in your latency budget.
Factor | /search | /research |
Latency | Under one second | Multi-step, seconds to minutes |
Sources synthesised | One to two pages | Many, across multiple steps |
Benchmark | 92% F-score, Verified SimpleQA | 61% accuracy, SealQA-0 |
Best for | Live chat, agent tool calls, RAG | Reports, due diligence, multi-hop questions |
Where it runs | User-facing path | Async job |
A practical pattern: route simple queries to /search by default, and escalate to /research when the question is flagged as multi-hop or when /search confidence is low. A [financial services] team building a research assistant uses /search for live lookups and /research for the deeper memos.
Why running both on one API matters for enterprise
Linkup is the only web search API for LLM teams that ships both a #1 sub-second /search and a #1 multi-step /research behind one contract, one SDK, and one security posture. You do not stitch together two vendors with two compliance reviews.
That posture includes SOC 2 Type II, Zero Data Retention by default, GDPR compliance, and Bring Your Own Cloud (BYOC), so queries never leave your VPC. Both endpoints are reachable through the OpenAI SDK, LangChain, LlamaIndex, CrewAI, Vercel AI SDK, n8n, and an MCP server. One integration covers fast lookups and deep research without a second procurement cycle.
Start with /search for anything user-facing and reach for /research when the question needs synthesis. Both are documented at https://docs.linkup.so/pages/documentation/get-started/introduction, or create a key at https://app.linkup.so/sign-up to test them against your own queries.
FAQ
What is the difference between /search and /research in Linkup?
/search is a single-pass web search that returns answers in under a second. /research runs multiple search steps to synthesise answers across many sources for harder, multi-hop questions.
Which Linkup endpoint is the best search API for low latency?
/search is built for low latency, returning results in under one second and ranking #1 among sub-second web search APIs with a 92% F-score on Verified SimpleQA.
When should I use /research instead of /search?
Use /research when the answer requires combining facts from several sources, when the question is ambiguous and needs decomposition, or when accuracy on hard questions matters more than response time.
Can I use both endpoints with the same API key and SDK?
Yes. Both /search and /research are available through one contract, one SDK, and one security posture, accessible via the OpenAI SDK, LangChain, LlamaIndex, CrewAI, Vercel AI SDK, n8n, and MCP.
Is Linkup's web search API for LLM secure enough for enterprise?
Linkup is SOC 2 Type II certified, GDPR compliant, runs Zero Data Retention by default, and supports Bring Your Own Cloud so queries never leave your VPC.
