The Best Company Funding Data API in 2026: What an Independent Benchmark Found

*Firecrawl's two figures are the top score on each board, from different Spark models: 100% freshness (Spark 2) and 92.3% enrichment (Spark 1 Mini). The other rows show a single representative configuration. The full results table lists every arm.

You hand a provider a company domain and ask one thing: what is the most recent round this company raised, named correctly? GTM teams use the answer to prioritize outbound, VCs to source deals, RevOps to keep a CRM from going stale. Every vendor claims the best coverage, and until recently no one had tested that claim the same way across all of them.

In August 2026 the independent Openbenchmarks company funding board did. It runs 17 providers over the same company domains, judges each returned stage against a reviewed ground truth, and splits the score into two boards: freshness (rounds announced in the last 30 days) and enrichment (older rounds). Firecrawl's agent tops both. The rest of the field trades accuracy, speed, and cost in ways that make "best" depend entirely on which job you are running.

This post walks through the full results and how to pick a provider by that job. The benchmark is public and reproducible, so every number here is checkable, including where Firecrawl loses.

A company funding data API takes a company identifier, usually a domain, and returns structured facts about that company's funding. The headline field is the latest funding stage: Seed, Series A, Series B, and so on, plus the amount, the date, and sometimes the lead investor.

The hard part is not the schema. It is being right. A funding round is a moving target. It gets announced on a company newsroom, picked up by press, filed with regulators, and eventually lands in a structured database, and each of those steps adds lag. A provider that is perfect on rounds from two years ago can be badly wrong on a round from last Tuesday, and the reverse happens too.

Providers answer the question in three different ways, and the difference matters more than any single accuracy number:

• Long-running agent APIs dispatch an agent that researches each company live at request time, reading sources and deciding what the latest round was. Slowest and most expensive per call, priced on how much work they do.
• Web search APIs query a search index and extract the answer from what comes back. Fast and cheap per call, with accuracy that tracks how well the round was covered on the open web.
• GTM data providers return a stored record from a maintained company database. Predictable latency and per-record cost, with coverage set by the vendor's own ingestion rather than your query.

None of these is strictly better. They are different purchases for different jobs, which is exactly why a single ranked list of funding vendors is misleading.

The Openbenchmarks funding board holds everything constant except the provider. Every provider gets the same company domains through its documented endpoint, the natural-language arms get one shared instruction and the same output schema, and every returned stage is judged by the same policy against a reviewed ground truth. The ground truth itself is verified against company newsrooms, investor announcements, and regulatory filings before a round is included. That is the setup that makes the numbers comparable.

The board splits into two, and this split is the whole point:

• Freshness measures rounds announced in the trailing 30 days. It tells you how fast a provider indexes a new announcement.
• Enrichment measures rounds older than 30 days. It tells you how completely a provider has backfilled history.

These reward opposite things. A provider that indexes fast can still hold a thin historical record, and a deep historical database can lag badly on last week's news. Averaging them into one rank would describe neither, so they are ranked separately. Every score sits against that provider's own denominator, because vendors joined the benchmark at different times.

One more thing to keep in mind: only the latest stage is judged for correctness. Round amount, total raised, and round count are counted for presence, not accuracy. So a high "funding fields returned" number means a provider filled the fields in, not that the numbers in them are right.

Here is the full board, all 23 measured arms across the 17 providers, grouped by how each one produces its answer. Freshness is latest-stage accuracy on rounds announced in the trailing 30 days; enrichment is the same on rounds older than that. Latency and estimated cost are per company, as measured by the benchmark's own runner. The two reviewed dataset exports (Crunchbase, Harmonic) carry no latency or cost because they are not timed API calls.

Source: Openbenchmarks company funding board, freshness and enrichment boards, last run September 2026. Freshness and enrichment are latest-stage-correct percentages against each provider's own denominator. Cost is modeled from observed billing units at published rates, not invoiced.

Which provider is "best" flips depending on which board you need, so here is the honest reading of the board.

On recent rounds, the live-web providers win. The freshness board is topped by Firecrawl's agent at 100% latest-stage-correct, with Exa instant at 98.0%, Exa's agent at 97.0%, Firecrawl's second model at 96.0%, and Parallel's Task API at 95.0%. Crunchbase's export matches them at 95.1%. The stored-database GTM providers trail hard here: Apollo at 59.7%, People Data Labs at 13.3%, CompanyEnrich at 12.7%. That gap is the lag. A database has not ingested a round announced nine days ago; an agent reading the live web has.

On historical rounds, it evens out but the agents still lead. The enrichment board is topped by Firecrawl at 92.3%, Parallel at 90.0%, Exa's deep and agent modes near 88.6%, and Crunchbase at 85.8%. The strongest true-API GTM provider on this board is Fiber at 84.9%. The weaker database providers stay weak: Ocean.io at 5.5%, Explorium at 21.9%.

Cost and speed go the other way. The agents that lead on accuracy are the slowest and priciest per call, running to a minute or more per company. The web search APIs return in seconds and cents, and the GTM database lookups in a few hundred milliseconds. Latency and price are the axes where the stored-record providers earn their place despite the accuracy gap: for a known company where a recent round is not the concern, a database read is cheap and fast.

The rest of this post goes provider by provider, grouped by type.

Firecrawl is the web data API to search, scrape, and interact with the web at scale, and its /agent endpoint researches a company live to return its latest funding stage.

Benchmark result: the top score on both boards, 100% freshness (Spark 2) and 92.3% enrichment (Spark 1 Mini).

Firecrawl is the only provider to top both boards. On the freshness board, its agent running the spark-2 model returned the correct latest stage for 100% of measured companies, the highest score on the board, and it did so at an estimated $1.98 a run, the cheapest of the long-running agents. On the enrichment board, the spark-1-mini model posted 92.3% latest-stage-correct, again the top score, and returned a funding stage on 100% of companies. The reason is structural: funding rounds live on company newsrooms, press wires, and filings, which is exactly the unstructured web content Firecrawl was built to read. Nothing has to be pre-ingested for the agent to find it, which is why it catches last week's round that a stored database has not reached yet.

Firecrawl covers both speed and depth, and the trick is picking the right endpoint for the job. The /agent endpoint is the depth-first option: a live research run that leads both boards and is the most accurate way to answer "what did this company raise," ideal for high-value lookups and monitoring workflows where being right matters more than shaving milliseconds.

For volume, Firecrawl search is the fast path, and the benchmark does not even measure it: the board scores only /agent. Search resolves a query against the live web in a single call rather than dispatching a research agent, so it is quicker and cheaper per request than the agent while keeping the same live-web freshness advantage. Between the two endpoints, Firecrawl spans the accuracy leaders and the fast, low-cost tier from one API, so you rarely have to leave it to change the speed-accuracy tradeoff.

• /agent: Dispatches a research agent that navigates and gathers funding data wherever it lives on the web, returning a strict JSON schema you define. For the broader landscape of structured data extraction tools, that guide covers schema-first extraction across providers
• search: Finds fresh funding announcements and press coverage from the live web, returning full-page content in one call
• scrape: Turns a company newsroom, TechCrunch article, or investor page into token-efficient Markdown or structured JSON
• crawl: Extracts every page of a company's newsroom or investor section on a schedule
• parse: Converts regulatory filings and investor PDFs into text an agent can reason over
• interact: Handles logins and dynamic pages that a plain scrape cannot reach
• monitor: Watches company pages for change, so a new round shows up as an event you can act on

Install:

# Python SDK
pip install firecrawl-py
 
# CLI (scrapes, searches, and agent runs from the terminal)
npx -y firecrawl-cli@latest init --all --browser
 
# MCP server (plug into Claude, Cursor, or any MCP client)
https://mcp.firecrawl.dev/v2/mcp-oauth

Example:

from firecrawl import Firecrawl
from pydantic import BaseModel, Field
from typing import Optional
 
firecrawl = Firecrawl(api_key="fc-...")
 
class FundingRound(BaseModel):
    latest_stage: str = Field(description="Most recent round, e.g. Series B")
    amount_raised: Optional[str] = Field(None, description="Amount of the latest round")
    announced_date: Optional[str] = Field(None, description="Date the round was announced")
    lead_investor: Optional[str] = Field(None, description="Lead investor on the round")
 
# Research a company's latest funding round with the agent
result = firecrawl.agent(
    prompt="Find the latest funding round for example-company.com: stage, amount, date, and lead investor.",
    schema=FundingRound,
    model="spark-2",
    max_credits=100,
)
 
print(result.data)

Honest take: If your job is being right about a company's funding, the benchmark makes the case for you: Firecrawl's agent is the most accurate option measured and the only one to top both boards, at the lowest agent cost on the freshness board. Reach for /agent when accuracy is the priority, and drop to search when you are enriching at volume and want speed and lower cost, both from the same API. With Firecrawl Keyless, search works without an API key: every developer gets 1,000 free credits a month with no key and no signup, which is enough to test funding enrichment against your own company list before committing.

Cons: The agent is a research run, so its per-call latency and cost sit above a plain search or database lookup, which is why search exists for the high-volume path. Firecrawl also holds no firmographic database of its own, so it will not return the stable non-funding fields (employee count, headquarters) that a GTM provider stores alongside the round.

Beyond the agent, Firecrawl has an MCP server you can plug into Claude, Cursor, or any MCP client, and a CLI for running scrapes and searches from the terminal. Full reference at firecrawl.dev/agent, launch details in introducing /agent, and the finance use case breakdown at firecrawl.dev/use-cases/finance-teams.

Parallel researches a company live through its Task API and posted 95.0% latest-stage-correct on freshness and 90.0% on enrichment, a few points behind Firecrawl on both. It also runs a faster index-backed Responses arm that trades some depth for lower latency, so you can move between the two without switching vendors.

Best for: balanced accuracy across recent and historical rounds from a single research API.

Tradeoffs: priced per completed run rather than per record, so it costs more than the cheap search tier for bulk work. See the head-to-head Firecrawl vs Parallel comparison, and the deep research APIs guide for the wider pattern.

Exa is a search engine built for AI, and its instant search is the fast, cheap end of the board: about 1.6 seconds a call at a $0.007 per-search list price, hitting 98.0% freshness and 87.0% enrichment. The $0.70 shown for Exa instant in the results table is the modeled all-in cost per company, which bundles the model tokens each run spends on top of that search price, so it is not comparable to the $0.007 unit rate. Its deep-reasoning and Agent arms add a little historical accuracy at the cost of the speed that makes instant interesting.

Best for: enriching a large list where latency and per-call cost matter more than the last few points of accuracy.

Tradeoffs: instant mode trails the research agents on the hardest historical rounds, and coverage tracks open-web presence. See the head-to-head Firecrawl vs Exa comparison for the full breakdown.

Nimble's standard search depth landed in the low-80s on both boards (84.8% freshness, 81.3% enrichment), below Exa and the research agents but ahead of the weaker databases. The depth setting matters: its lite tier drops into the 60s, so it is not the one to reach for on funding. For how it compares across scraping and search workloads, see our Nimble alternatives comparison.

Seltz markets a companies-scoped search, but on funding it scored low across both of its scopes (22.0% freshness on the companies scope, 71.0% on news) and does not disclose pricing, which makes an accuracy-per-dollar read impossible. On these numbers it is not a strong fit for latest-round data.

The firmographic vendors answer from a stored record instead of the live web. That buys predictable latency, a few hundred milliseconds per lookup, and predictable per-record cost. What it costs is freshness: a database has not ingested a round announced last week, which is why the whole group trails on the freshness board and closes the gap only on historical enrichment. Here they are ranked by enrichment accuracy:

Crunchbase and Harmonic are measured from reviewed exports, not timed API calls, so they carry no latency or cost.

The short read: for a known company where a recent round is not the point, a stored-record lookup is faster and cheaper than a research run, and among these Crunchbase's export and Fiber hold up best on accuracy. Below them the numbers fall off quickly, and several well-known enrichment platforms score poorly on this one field, a reminder that being good at contact data says nothing about being good at funding rounds. If you are scraping funding data off the web instead of buying it, our Crunchbase scraping guide walks through the build with Firecrawl.

Start from the job, not the top of a ranking. The benchmark makes four workflows cleanly separable:

1. Real-time funding alerts. You want to catch a round in the days after it is announced. Read the freshness board only, where the live-web providers lead and Firecrawl's agent tops it at 100%. A stored database will lag by definition here.
2. CRM backfill enrichment. You are filling in historical rounds on known companies and speed does not matter. Read the enrichment board with cost in mind. Firecrawl and Parallel are the most accurate, Crunchbase if you have a license, and Fiber if you want a fast stored-record API.
3. Enrichment at scale. You have a large list and price per correct answer is the constraint. Compare accuracy-per-dollar across the fast, single-call endpoints: Firecrawl search, the web search APIs, and the GTM databases. A long-running research run is overkill per row here, which is exactly why Firecrawl also exposes search for this path.
4. Live in-product enrichment. You need an answer inside a user-facing flow where latency shows. Read median latency first among the web search APIs and GTM databases; the long-running agents are too slow for this seat.

Then run the benchmark yourself. The Openbenchmarks harness, the frozen inputs, and the evaluation code are public, so you can swap in your own company list and providers and see how the ranking holds on your cohort rather than theirs.

If your workflow is discovery and freshness rather than a domain-keyed database read, Firecrawl's agent is the highest-leverage addition to the stack: it topped both funding boards by reading the live web the way the ground-truth reviewers did. For the broader picture of how these pieces fit together, the best investment research APIs guide covers the market and fundamentals layers that sit alongside funding data, the best AI sales automation tools for GTM roundup covers where funding signals plug into revenue workflows, and the AI agents guide covers how to wire a research agent into a pipeline that decides for itself which companies to enrich.