Firecrawl Review (September 2026): Alexandria, the $75M Series B, and Whether the 4.6 Rating Holds

On September 22, Firecrawl announced a $75M Series B and, in the same post, launched Alexandria, a layer that lets an agent query paid data providers and Firecrawl's own indexes through the API it already calls (Firecrawl announcement, Sept 22, 2026). Our September 4 review rated Firecrawl 4.6 and saw neither event.

The rating holds at 4.6. Alexandria is a real new capability for agents that need structured data from many sources, and the plan prices we track did not move after the raise. What keeps the score from climbing is billing: the pricing page carries no Alexandria row, because every Alexandria tool has its own price. So try it inside your current plan, set a per-key credit cap first, and read each tool's price before your agent calls it.

We pulled the pricing page and the Alexandria docs on September 28 and read them against the announcement and two outside reports. One correction to our own earlier numbers came out of that, covered below.

What changed on September 22

The round was led by Smash Capital, with Altos Ventures, Nexus Venture Partners, Y Combinator, Freestyle, and Offline Ventures participating, per the Firecrawl announcement. SiliconANGLE's report the same day matches the amount and the participating investors. No valuation was disclosed, per RuntimeWire.

Firecrawl says it will spend "a good chunk" of the money buying knowledge from people. It already pays official providers through individual agreements, Wikimedia Enterprise being the named example, and plans a self-service system for individuals and organizations to earn when agents use their data (announcement). That system is a provider waitlist on the Alexandria page today. Developers can use Alexandria now; publishers cannot yet sign up and get paid.

How Alexandria works: discover, inspect, run

Scrape and search answer "what does this URL say." Alexandria answers a different question: which source has the data at all. An agent sends a search with alexandria in its sources and gets back ranked tools alongside web pages. Each tool lists its provider, capability, inputs, response contract, and price. Next, the agent calls the chosen tool through the same scrape endpoint, with no page URL (Alexandria docs).

// Adapted from the Alexandria docs. Discovery is free when sources is only "alexandria".
const found = await firecrawl.search("podcast conversations about AI agents", {
  sources: ["alexandria"],
  limit: 2,
});
// found.tools: provider, capability, inputs, response contract, price

// Execution bills at the tool's listed price.
const episodes = await firecrawl.scrape({
  alexandria: {
    provider: "particle",
    capability: "podcasts/episodes/search",
    options: { semantic_search: "AI agents", limit: 2 },
  },
});

On September 28 the product page listed 93 providers and 640 capabilities across 20 categories, plus 113M+ indexed papers, READMEs, issues, and docs. Named providers include Fiscal.ai for company financials, FRED and the World Bank for economic data, Benzinga for financial news, and Particle for podcast transcripts. Firecrawl's own entries sit next to them: the Research Index (43M abstracts), the Developer Index (70M+ repo artifacts), and a Government Index for laws, tax forms, and permits. Provider counts vary by source; RuntimeWire, citing Firecrawl's launch post on X, reported more than 100. Treat any count as approximate and growing.

Firecrawl's announcement uses startup research as its example: query the directories that list whole startup batches, compare what each company builds, then run people enrichment to find a buyer at the companies that match. The agent answers both questions over the same connection, with no new integration for the second one.

The 21% number is Firecrawl's own

Firecrawl reports that agents using Alexandria scored 21% higher on answer quality than agents using built-in web tools, with the same model and prompts across 845 tasks and blind AI judging (announcement). That is the vendor's internal evaluation. No outside party ran it, and the task set is not published.

RuntimeWire's account describes it as roughly 1,000 catalogue-based questions across more than 10 categories and notes that the questions were designed around catalogue data. That design favors Alexandria by construction. Read the 21% as evidence that Alexandria helps when the answer lives with a provider Firecrawl has connected. It says little about agents whose answers sit on the open web.

What Alexandria costs

The pricing page (fetched 2026-09-28, stamped effective September 4, 2026) kept the same plans, and its credit table has no Alexandria row. Per-endpoint rates are unchanged: Scrape and Crawl at 1 credit per page, Monitor at 1 credit per page per check, Search at 2 credits per 10 results, Interact at 2 per browser minute.

The missing row does not mean Alexandria is free. The Alexandria docs spell out the model. Discovery costs nothing if your search uses only the alexandria source. Add web to the same search and the web results bill at the normal Search rate. Executing a tool is "charged at the tool's listed price," drawn from your plan credits. So you can budget one call precisely, by reading the price field before running it. What you cannot do is budget a month from the pricing page, because the prices live on 640 capabilities, not in one table.

For an agent that picks tools on its own, that is a real risk. Firecrawl already ships the fix. Its billing docs let you set a credit limit per API key, per day, week, or month. Give the agent that uses Alexandria its own key with a cap, and set a monthly pay-as-you-go limit in billing settings so a run of expensive tool calls stops at a number you picked (pricing FAQ).

One more setup step catches teams. Some providers require an org admin to accept third-party data terms. Calls to those providers return an error with a link to the terms, agents need explicit user authorization before accepting, and Agent runs skip those providers until someone does (Alexandria docs). If an agent's answer looks thin, check the skipped-providers list first.

A correction to our September 4 numbers

Our September 4 review listed 50 concurrent requests on Standard, 100 on Growth, and 150 on Scale. The current pricing page lists 25, 50, and 100 concurrent browsers, and it carries an effective date of September 4, 2026, the same day that review published. Whatever the earlier render showed, plan against the current figures. That moves the plan picks: a team that sized Standard for 50 parallel renders gets half that, and Growth is now the tier for that load.

The same page settles the other open item from September 4. The Hobby and Standard dollar amounts, which the slider rendered blank then, now read $16 and $83 a month billed yearly on the Markdown pricing page, matching July.

Does the funding change the buy decision

A big round can signal churn for buyers: price hikes, a pivot, a product freeze ahead of a sale. None of that shows up in the record. On September 28, six days after the round, the pricing page still listed Hobby at $16 and Standard at $83 billed yearly (the July figures) and Scale at $599, the figure our September 4 review confirmed. The founders, Caleb Peffer, Nicolas Silberstein Camara, and Eric Ciarla, are unchanged, per RuntimeWire. Firecrawl's stated uses for the money are provider payouts, better search, and deeper indexes, which extend the product buyers already pay for (announcement).

One loose thread. RuntimeWire found a Form D showing $82,063,463 of preferred stock sold, first sale August 31, about $7.1 million more than the announced round, and neither disclosure explains the gap. It does not change what you pay or what the API does.

Rating: 4.6 holds

Alexandria earns a place in the product. For an agent whose job is research across company data, filings, government records, and developer docs, one connection with inspectable per-call prices replaces a stack of separate integrations. Plan prices held after the round, and pay-as-you-go plus per-key limits give buyers the controls that per-tool pricing needs.

Three things keep the score at 4.6 instead of moving it up. Firecrawl's 21% is the only quality number available, and it comes from questions built around its own catalogue (RuntimeWire). Budgeting Alexandria means inspecting tools one at a time, since the pricing page lists none of them. And the concurrency correction means Standard does less parallel work than our last review told readers.

The verdict

If you already pay for Firecrawl, try Alexandria inside your current plan. Browsing the catalogue costs nothing, so start with a tools-only search, read the price on the two or three tools your agent actually needs, and put that agent on its own API key with a credit limit before it runs unattended. If you planned capacity around 50 concurrent browsers on Standard, redo it at 25.

If you are evaluating Firecrawl fresh, the tier advice from September 4 still applies: start on Free or Hobby and move to Standard once monthly burn clears about 6,000 credits, since Hobby includes 5,000 per the pricing page. Alexandria is a reason to pick Firecrawl over Apify or ScrapingBee only if your agent has to find which of dozens of sources holds the answer. For scraping a known list of URLs it adds nothing over /scrape, and our Firecrawl vs Apify vs Browse AI comparison still decides that call. For multi-source research agents, start on Firecrawl here and give Alexandria a capped key. Rated 4.6 of 5.