List building · Updated
Parallel review (2026): strong web research, the wrong tool for big lists
Parallel is one of the best APIs for researching the companies on a list you already have: accurate, cited answers in seconds to a minute. It is a poor tool for building new lists, the slowest we tested with the fewest rows. Use it to research a list, not to build one.
pipe0 buys from Parallel and runs its web search, page extraction and company entity search on it. The scores come from our own benchmark, not from that relationship. How we review
What we reviewed
Parallel builds web search and research APIs for AI agents. We reviewed it for the two GTM jobs teams bring to us: researching the rows on a list, and building a new list from a prompt.
Tested for
- Finding a specific page or fact per company with Search and Task
- Local business lists with Task pro and ultra, FindAll and Entity Search
- Speed and cost against Exa, DataForSEO, Crustdata and Amplemarket
Not scored here
- Monitor, Chat and the Responses API
- Extract at volume as a crawler
Our score for Parallel, the average of four criteria
Excellent for researching rows you already have: accurate, cited answers in seconds to a minute. For building new lists it was the slowest tool we tested and returned the fewest rows.
Best for
- Researching one fact or page per company on a list you already have
- Agents that need fast, cited web search
- Qualifying companies quickly with Entity Search
Skip it if
- You want to build a new list of hundreds or thousands of businesses
- You need addresses and phone numbers for local businesses
- The step has to finish in under a few minutes
Clear APIs with structured output and a citation and confidence per answer. Picking among nine Task processors takes some reading.
Research answers were accurate: Task base solved 5 of 5 hard lookups. Lists were short and loose, 17 to 200 rows, and 55% of NYC rows were inside the city.
Search at $5 per 1,000 requests and Task base at $0.01 a run are fair for research. Verified list rows from FindAll cost us $48 per 1,000.
A free hosted Search MCP, a Task MCP, citations on every answer, and 600 searches a minute by default.
Parallel at a glance
- APIs
- Search, Extract, Task, FindAll, Entity Search, Monitor and Chat Source
- Task latency
- Lite 10 to 60 s, base 15 to 100 s, pro 2 to 10 min, ultra 5 to 25 min, per Parallel's docs Source
- Rate limits
- Search, Extract and Entity Search 600 a minute; Task 2,000 a minute Source
- MCP servers
- Search MCP at search.parallel.ai/mcp, free without a key; Task MCP with authentication Source
- Search controls
- Search objective plus keyword queries, domain allow and block lists, and an after-date filter
What is Parallel?
Parallel Web Systems builds web APIs for AI agents. Search returns ranked excerpts for a stated objective, Extract reads pages, Task answers a research question with a structured and cited result, FindAll builds verified lists, and Entity Search finds companies.
What is Parallel best at?
Researching the rows you already have. Most GTM research is one question per company: who runs it, where its pricing page is, which tools it uses. On five company websites, including one with almost no search footprint, Parallel's Search ranked the right page first on four, in one to three seconds. The base Task processor answered all five, including a small company's imprint page at an unguessable path, in 22 to 63 seconds for a cent each.
Every Task answer carries its sources, the reasoning and a confidence level, so a person or an agent can check it.
Which Parallel processor should you use?
Base for research questions; skip lite and the fast search modes for facts. In our tests lite missed exactly the hard case base solved while saving little time, and turbo search answered a company revenue question with the company's acquisition price. Fast modes are for finding pages, not for reading facts off them.
Can Parallel build lead lists?
Not well. Task runs returned 17 to 200 local businesses per list and took eight to fourteen minutes, slower than any other tool we compared. Exa's agent returned more rows on all four lists in three to six minutes, and DataForSEO returned 826 to 3,959 in seconds.
How many businesses each tool finds
Unique rows per list, one run each. All four panels share one scale, so a short bar is a short list.
Barbershops in New York City
DataForSEO
Exa ultra
Exa medium
Parallel Task pro
Parallel Entity
Crustdata
Amplemarket
HVAC contractors in Dallas–Fort Worth
DataForSEO
Exa ultra
Exa medium
Parallel Task pro
Parallel Entity
Crustdata
Corporate event venues in Chicago
DataForSEO
Exa ultra
Exa medium
Parallel Task pro
Parallel Entity
Crustdata
Amplemarket
Property managers in Miami
DataForSEO
Exa ultra
Exa medium
Parallel Task pro
Parallel Entity
Crustdata
Amplemarket
Company databases count every row they returned; most were neighbouring businesses such as suppliers, software and planners.
How long a list takes
Time per run, from each vendor's own timestamps. Each bar spans the fastest and the slowest list.
Filter and company search
DataForSEO
Filter search over map listings
Parallel Entity
Company search
Prompt-based agents
Exa medium
Prompt-based research agent
Exa ultra
Prompt-based research agent
Parallel FindAll
Verified list search
Parallel Task pro
Prompt-based research task
Parallel Task ultra
Prompt-based research task
The line marks 10 minutes. Past it, a run doesn't fit an interactive workflow or the step limit most automation tools set.
Higher processors don't fix it. The ultra processor returned fewer New York City barbershops than pro, and only 55% of them were in the city.
What do FindAll and Entity Search cost per row?
They sit at opposite ends. FindAll verifies every match, so its rows are accurate and expensive: 24 Chicago event venues cost $1.15, about $48 per 1,000. Entity Search returns 81 to 158 companies in two to five seconds for under a cent, but the rows are company profiles without an address or a phone number, and it stops well short of large lists.
What 1,000 rows cost
Vendor price per 1,000 rows, before any reseller margin. Log scale: each tick is ten times the last.
Filter and company search
DataForSEO
Filter search over map listings
Parallel Entity
Company search
Prompt-based agents
Exa medium
Prompt-based research agent
Exa ultra
Prompt-based research agent
Parallel FindAll
Verified list search
Parallel Task pro
Prompt-based research task
Parallel Task ultra
Prompt-based research task
Per-run pricing makes a short list expensive per row. Parallel Entity is the cheapest per row, but its rows are company profiles without an address or a phone.
Parallel and pipe0
pipe0 runs its web search, page extraction and company entity search on Parallel, and its research agent uses the Task API for questions that need a deeper look. New local and company lists go to DataForSEO, the company databases or Exa instead.
Parallel, Exa or DataForSEO: which should you use?
Use Parallel to research a page, a fact or a person per company, with a citation you can check. Use DataForSEO, a company database or Exa to build the list itself.
Which tool for which list
Every cell cites the measurement behind it. Company databases are Crustdata and Amplemarket, searched with their filters.
| Job | DataForSEO | Exa | Parallel | Company databases |
|---|---|---|---|---|
| Local businesses by categoryBarbershops in New York City | Best pick826 to 3,959 rows per city in seconds | Works146 to 341 rows in 3 to 6 minutes | Weak17 to 200 rows in 8 to 14 minutes | Weak26 barbershops, most of them noise |
| Phone, hours and rating per locationHVAC contractors to call | Best pick80 to 96% of rows with a phone | Mixed97% with a phone in Dallas, 8% in NYC | WeakEntity Search returns no phone | Not built for itNo phone per location |
| Companies by firmographics or tech stackUS apparel brands on Shopify | Not built for itOnly businesses with a map listing | Weak26 brands, 25 already in Crustdata | Not testedEntity Search wasn't run on this | Best pick77 verified Shopify brands in seconds |
| Lists published on the webExhibitors at a trade show | Not built for itNo event or membership data | WorksA few hundred accurate rows per run, never the full list | Not testedFindAll fits the shape; not run | Not built for itNo exhibited-at filter |
| One researched answer per rowThe imprint page of a small company | Not built for itNot a web research tool | Not testedNot benchmarked for this | Best pickTask base solved 5 of 5, in 22 to 63 s | Not built for itNot a web research tool |
One run per scenario. “Not tested” means we didn't measure it, not that it fails.
Parallel pros and cons
Pros
Finds the right page
Searching a company's own domain, Parallel ranked the right page first on 4 of 5 sites, in one to three seconds.
Reliable research tasks
The base Task processor solved 5 of 5 hard lookups, including a small company's imprint page, in 22 to 63 seconds for $0.01 each.
Citations by default
Every Task answer comes with its sources, the reasoning and a confidence level, so a person or an agent can check it.
Fast company search
Entity Search returned 81 to 158 companies in two to five seconds for under a cent a call.
Cons
Slow lists
Task pro and ultra took eight to fourteen minutes per list, too long for an interactive step or a 10-minute limit.
Short lists
Tasks returned 17 to 200 local businesses where DataForSEO returned 826 to 3,959, and ultra found fewer NYC barbershops than pro.
Expensive verified matches
FindAll's matches were all verified, but 24 Chicago venues cost us $1.15, about $48 per 1,000.
Fast modes cut corners
Turbo search answered a revenue question with an acquisition price. Pick the processor and mode for the question.
Parallel pricing
Parallel has a free plan. Paid plans start at Free pay as you go, with $5 in free credit each month.
| Plan | Price | What you get |
|---|---|---|
| Search | $5 per 1,000 requests | Basic and advanced modes, 10 results; turbo and fast $1 per 1,000. |
| Extract | $1 per 1,000 URLs | Page contents as excerpts or full markdown. |
| Task | $5 to $300 per 1,000 runs | Lite $5, base $10, core $25, pro $100, ultra $300, with larger ultra tiers above. |
| FindAll | From $0.25 per run | Base: $0.25 per run plus $0.03 per match. Core and pro cost more per match. |
| Entity Search | $5 per 1,000 requests | Up to 100 companies per request; beta. |
- The $5 monthly credit needs a card on file.
- Task prices are per successful run.
Prices as of Oct 15, 2026, from Parallel's pricing page.
Parallel alternatives
| Alternative | Pick it when |
|---|---|
| pipe0 company entity search | You want Parallel's Entity Search in a sheet, followed by enrichment pipes on the companies it finds. |
| Exa | You want a prompt-based list of a few hundred rows in minutes, not a quarter of an hour. |
| DataForSEO | Your list is local businesses: thousands of rows per city in seconds. |
| Crustdata | Filters can describe your companies, from headcount to tech stack. |
Frequently asked questions
What is Parallel Web Systems?
Parallel builds web APIs for AI agents: Search, Extract for page contents, Task for research questions with cited answers, FindAll for verified lists, and Entity Search for companies.
Is Parallel good for building lead lists?
Not in our tests. On four local business lists, Parallel's Task API returned 17 to 200 rows in eight to fourteen minutes. Exa returned more every time, and DataForSEO returned 826 to 3,959 in seconds.
What is Parallel best at?
Research on rows you already have. Searching within a company's domain, it ranked the right page first on 4 of 5 sites in one to three seconds, and its base Task processor solved 5 of 5 hard lookups.
How much does the Parallel Task API cost?
From $5 per 1,000 runs for lite to $300 for ultra, with base at $10. Larger ultra tiers cost up to $2,400 per 1,000.
Exa or Parallel?
For building lists, Exa: it returned more rows in less time in all four of our local scenarios. For cited research answers per row, Parallel's Task API was the most reliable we measured; we didn't benchmark Exa on that job.
Next-gen enrichment & search.
Build revenue systems that scale. For humans, agents, and apps. Replace tools like Clay, n8n, Hightouch, etc.
Try pipe0