Firecrawl Extract Pricing: Skip the Dual Bill

By the fastCRW team · fastCRW credit costs verified against the canonical fact sheet 2026-05-18 · Firecrawl Extract pricing is reportedly/HN-sourced and dated 2026-05-18 — re-verify against firecrawl.dev/pricing before relying on any dollar figure · fastCRW launch pricing expires 2026-06-01 · Verify independently before buying.

Disclosure: We build fastCRW, so this is a vendor-authored comparison — weight it accordingly. Only the fastCRW side of the cost math is frozen and source-verifiable; the Firecrawl Extract pricing here is reported, not locked, and we flag every figure as such. We also include a section on where Firecrawl genuinely wins.

Firecrawl Extract pricing: why it can be two bills, not one

Firecrawl Extract pricing is the part of the bill that surprises people, because the AI extraction endpoint reportedly does not draw from the same credit pool as your scrape and crawl calls. The pattern that gets reported on HN and Reddit (competitor-profiling.md, 2026-05-18 — reportedly, not an S0-locked figure) is a separate token-based subscription stacked on top of whatever scrape plan you already pay for. So a single workload — "scrape the page, then pull structured fields out of it" — can land as two line items on the invoice instead of one.

That is the whole reason this page exists. If you are evaluating /v1/extract on Firecrawl, the question is not just "what does an extract call cost," it is "what does my all-in monthly cost look like once the extract subscription sits on top of the scrape plan." Below we walk the reported dual-billing pattern, attach honest provenance to every dollar figure, and then show fastCRW's flat side — the only part of this comparison that is frozen and traceable to a single source.

Scrape and crawl credits vs the separate Extract token plan

On Firecrawl, base scrape and crawl are metered in credits at 1 credit per page — that base rate is the one Firecrawl number locked in our price file (competitor-prices.lock.md, verified 2026-05-18). AI Extract is the part that is not in that lock. The reported model is that Extract runs as its own token plan: you buy extraction capacity separately, and it depletes on a different meter than the page credits you bought for scraping and crawling.

Why a Standard plan can mean two line items

The practical consequence: paying for a Standard-tier scrape plan does not "include" meaningful AI Extract volume the way a single per-page credit would. You add the extract subscription on top. For a team that extracts structured JSON on most pages — which describes most agent and enrichment pipelines — the extract plan is not an occasional add-on, it is a second recurring bill that scales with how much structure you pull.

Where the dual-billing surprise shows up in production

The trap is forecasting. You size the scrape plan from page volume, ship to production, and then the extract meter moves independently of the page meter. When extraction volume climbs faster than raw page volume — common as an agent starts pulling more fields per page — the bill stops tracking the one number you were watching. That decoupling, not the headline tier price, is what makes the all-in cost hard to predict.

The reported all-in cost of Firecrawl Extract

Here is the figure people search for, with the honesty it requires: a Standard scrape plan plus an extract add-on is reportedly ~$172–188/mo minimum (competitor-profiling.md, 2026-05-18 — HN/Reddit-sourced, "reportedly," and explicitly not in our S0 price lock). Treat that as an illustrative anchor, not a frozen fact. Firecrawl's published pricing changes, the extract line is the volatile part, and we have not independently re-verified the dollar total — so the instruction is the same one we give ourselves: check firecrawl.dev/pricing before you rely on any specific number.

Why the pattern, not the exact dollar, is the takeaway

The exact total is the part most likely to be stale by the time you read this. The pattern is durable and is what actually matters for a buying decision: extraction is billed on a separate meter from scraping, so your monthly cost is the sum of two independently-moving subscriptions. Whether that sum is $172, $188, or something else next quarter, the shape — two bills for one logical workload — is the thing to evaluate.

What scales the bill super-linearly with agent volume

When an agent does more work, it tends to extract more fields per page, not just visit more pages. So extraction volume can grow faster than page volume. On a model where extraction has its own meter, that means the second bill can outpace the first as the agent matures — the cost grows on a different curve than the page count you originally budgeted from. A single per-page credit, by contrast, only grows with pages.

How fastCRW prices structured extraction

This is the frozen side of the comparison — every number here traces to the canonical fact sheet. fastCRW charges a flat 5 credits for any request with formats: ["json"]. That is the whole rule. There is no separate extraction subscription on any tier — extraction is one operation on the same credit meter as everything else.

One flat 5-credit cost for any JSON request

Mechanically, JSON extraction is formats: ["json"] plus a jsonSchema on /v1/scrape. On the managed cloud there is also a convenience endpoint, /v1/extract — a 5-credit wrapper over the same /v1/scrape + formats: ["json"] path. Either way the cost is the same flat 5 credits, drawn from the one credit balance you already bought. There is no second meter to watch and no token plan to size. For the live tier table and what 5 credits maps to in dollars on each plan, see /pricing — we do not hard-code tier prices here because they change (and launch pricing reverts on 2026-06-01).

No separate extraction subscription on any tier

Because extraction is metered in the same credits as scrape, crawl, search, and map, your monthly bill is a function of one number — total credits consumed — not the sum of two subscriptions. That is the difference that makes forecasting tractable: you estimate credits from your operation mix and your plan allowance answers whether it fits.

Self-host the same JSON-schema extraction for $0

fastCRW's engine is AGPL-3.0 and self-hostable as a single static Rust binary. Self-hosted, structured extraction via /v1/scrape + jsonSchema costs $0 per 1,000 scrapes — you pay only for your own server. There is no managed cloud and therefore no extraction line item at all; the worst-case cost has a real floor, which a hosted-only metered model structurally cannot offer.

Honest limits: where fastCRW extraction stops

The flat-credit story has real edges, and a vendor comparison that hides them is useless. State these plainly:

• Single-URL only. There is no multi-URL batched /v1/extract the way Firecrawl Cloud offers. To extract across many URLs you iterate /v1/scrape concurrently or run a /v1/crawl. If your workflow is "give me one endpoint that fans out across a list of URLs and returns structured records," that batch convenience is something Firecrawl has and fastCRW does not.
• LLM providers limited to OpenAI and Anthropic. For formats: ["json"] extraction, the supported model providers are OpenAI and Anthropic only. (Note this differs from managed search answer mode, which uses a managed LLM — extraction and search have different provider surfaces.)
• No screenshot output. A request for formats: ["screenshot"] returns HTTP 422. If your extraction step also needs a rendered screenshot artifact, fastCRW will not produce one.

Where Firecrawl Extract genuinely wins

Even though the dual-billing shape is the cost downside, Firecrawl's extract surface has real strengths that a single flat credit does not replicate:

• Batched multi-URL extraction. One call that fans out across a URL list and returns structured records is a genuine convenience for large extraction jobs — you orchestrate that yourself on fastCRW.
• Provider and feature breadth. Firecrawl's cloud extract sits inside a broader, more mature feature surface (agentic and research endpoints, heavier anti-bot paths). If your pipeline depends on those, the second bill may be earning its keep.
• Ecosystem gravity. More tutorials, more community examples, more "this is what we used" posts. That is a real, if discovery-time, advantage.

Which extraction pricing model fits your workload

Map the decision to your traffic shape rather than the headline price:

The honest summary: if you genuinely use batched multi-URL extraction and the cloud-only extras, Firecrawl's separate extract plan can be worth paying for. If your extraction is single-URL and you want one predictable number on the invoice, a flat per-call credit is the more forecastable shape — and self-hosting takes the worst case to your server cost. For a broader cross-vendor look at how JSON-schema extraction is priced, see the structured extraction API pricing face-off.

Sources

• fastCRW credit costs, endpoints, gaps, self-host cost: live tiers at /pricing (derived from PLAN_DISPLAY).
• Firecrawl Extract dual-billing and ~$172–188/mo: marketing/competitor-profiling.md — reportedly, HN/Reddit-sourced, dated 2026-05-18, not in the S0 price lock. Re-verify at firecrawl.dev/pricing and docs.firecrawl.dev.
• Firecrawl base scrape/crawl 1 credit/page (the only locked Firecrawl figure): marketing/competitor-prices.lock.md, verified 2026-05-18.
• fastCRW repo and engine: github.com/us/crw · fastcrw.com.

Related: Firecrawl extract endpoint deep dive · Structured extraction with JSON schema · Firecrawl pricing explained · Structured extraction API pricing face-off