Firecrawl Pricing in 2026: How Much Does it Really Cost?

Key takeaways

• Firecrawl’s published pricing starts at $0 and reaches a $599/month equivalent on Scale, with Hobby, Standard, Growth, and Scale prices shown when billed yearly. Enterprise pricing is custom.
• A plain Scrape, Crawl, or Map starts at 1 credit per page, but the effective cost changes with the job. Search costs 2 credits per 10 results, AI-prompted Interact costs 7 credits per session minute plus the underlying Scrape, and advanced Scrape options can add credits per page.
• Firecrawl does not currently offer general pay-as-you-go billing beyond its free allowance. Self-serve credits also do not roll over, although rollover is available on Scale and Enterprise.
• TinyFish, a Firecrawl alternative, uses a different cost model. Search and Fetch cost 0 credits, Agent has a fixed cost of $0.016 per step, and Browser costs $0.002 per minute with no subscription. TinyFish also offers up to $13 in free sign up cash to test and build around.

Firecrawl pricing looks simple until your workflow does more than a plain scrape.

As of August 2026, its published plans run from free to a $599/month equivalent when billed yearly, but the plan fee is only the starting point. Search, browser interaction, advanced extraction, monitoring, and Agent work consume credits differently, so two teams on the same plan can burn through their allowance at very different rates and quickly rack up overage costs.

This guide breaks down every Firecrawl pricing plan, what each endpoint actually costs, where extra credits enter the bill, who each tier fits, what users report in practice, and how TinyFish compares for teams that want a different cost model.

How much does Firecrawl cost?

Firecrawl has a free tier, four paid plans with published prices, and a custom Enterprise tier. The paid prices below are the monthly equivalents currently displayed when billed yearly, so check the live rate before budgeting around a different billing cadence.

For a basic Scrape, one credit maps neatly to one page. However, the relationship stops being neat once you turn on other endpoints or output modes.

Pricing accurate as of 12th of August 2026. Firecrawl's rates change often, so check the live pricing page before budgeting.

Firecrawl’s Scrape options can add another layer. JSON extraction, Question, Highlights, enhanced proxy, and PII redaction each add 4 credits per page, while PDF parsing adds 1 credit per PDF page.

We’ll break down what that does to the effective cost of a page below.

How TinyFish pricing compares

TinyFish pricing starts from a different idea: reading the web should not cost the same as operating it.

Search and Fetch cost $0 on every plan. Agent costs $0.016 per completed step, while Browser costs $0.002 per minute. There are no subscriptions, so a team with irregular usage does not have to size a monthly credit bundle in advance.

The money-only differences are:

• Search and Fetch do not cost anything. Firecrawl charges 2 credits per 10 Search results and starts page retrieval at 1 credit per page, with some output modes adding more. TinyFish Search is $0 per request and Fetch is $0 per URL.
• Agent has one step-based meter. One Agent step and includes model inference, the stealth browser, anti-bot handling, and Agent proxy. Vault + Profile are platform capabilities rather than separate usage meters. Everything is included in one price and not as add-ons.
• Browser has its own time meter. TinyFish Browser charges $0.002 per minute, with a 60-minute session cap ($0.12). Browser proxy is metered separately rather than being folded into that browser-time cost.
• Failed runs cost $0. You only pay for successful operations.
• PAYG gives spiky workloads another option. Firecrawl’s self-serve plans require a recurring tier once the free allowance is no longer enough. TinyFish allows users to only pay for what they use with no monthly credit bundle.

For a simple illustration, if a successful Agent run happens to take 15 steps, it costs $0.24 at $0.016 per step. That is only arithmetic, not an estimate of how many steps a typical workflow should take. Actual Agent cost follows the number of completed steps in the run.

TinyFish also lets you set a spending cap in the dashboard (using max_steps) if you want an upper bound on spend per run.

Start building with TinyFish. Sign up for up to $13 in onboarding bonuses, no card required. Point it at a real target site so you can see what the meter looks like on your own workflow before committing.

Pricing accurate as of 12th of August 2026. Check the live pricing page before budgeting.

Model inference, the stealth browser, anti-bot, agent proxy, Vault and Profiles, and the SDKs, CLI, and MCP (Model Context Protocol) server are always included. Enterprise adds ISO 27001 certification, a research mode, and on-premise deployment.

Want to see a side by side comparison on your own query? The Search Arena runs your request across TinyFish, Exa, Tavily, and Parallel and compares cost, speed, and evidence quality.

Firecrawl Pricing Plans: A Breakdown

Firecrawl pricing is not one number. Your plan sets the size of the credit pool and how much work can run concurrently; the endpoint, browser time, extraction mode, and recurring checks determine how quickly that pool gets used.

Here is what each cost line actually meters.

Cost Line 1: Plan Tier, Credits, and Concurrency

Your Firecrawl tier controls two different production constraints: how many credits you reserve each month and how much work you can run at the same time.

Free starts with 1,000 monthly credits and 2 concurrent requests. Concurrency rises to 5 on Hobby, 50 on Standard, 100 on Growth, and 150 on Scale.

Above the free allowance, Firecrawl does not currently offer a general pay-per-use plan. You buy a recurring credit pool upfront, which is simple for steady workloads but means spiky teams have to estimate their future volume before choosing a tier.

For workloads that stay on the basic 1-credit-per-page path, the annual-billing prices give you a useful cost anchor:

That explains why Firecrawl can be economical for a predictable, high-volume crawl-to-content pipeline. The catch is that these numbers only hold when one page stays close to one credit. Turn on a 5-credit extraction mode, use AI-prompted browser interaction, or repeatedly monitor pages and the effective unit cost changes.

TinyFish separates basic reading from the metered operation layer. Search and Fetch are free, while Agent and Browser concurrency scales at a flat cost.

Before choosing a Firecrawl tier, model both monthly credit consumption and peak concurrency rather than assuming every URL will stay on the 1-credit path.

Cost Line 2: Scrape, Crawl, Map, Search, Interact, Monitor, and Browser

The second cost line is what you actually call. Firecrawl does not price every endpoint the same way.

Scrape, Crawl, and Map use 1 credit per page. Monitor also starts at 1 credit per page, but that charge applies per check, which changes the math for recurring workflows. Monitoring 1,000 pages once is one cost shape; rechecking the same 1,000 pages every day is another.

Firecrawl Search costs 2 credits per 10 results, rounded up. But that is only the search-result meter. If you ask Search to return scraped page content too, Firecrawl adds the normal Scrape cost for every returned page. A 10-result Search with a basic Scrape on all 10 results therefore consumes 12 credits before JSON, enhanced proxy, PDF parsing, or another advanced option enters the request.

Browser work splits again. Firecrawl’s Interact charges 2 credits per session minute when you supply the interaction code yourself, or 7 credits per session minute when AI prompts drive the interaction. The initial Scrape is billed separately. Its standalone Browser API is closer to a raw browser-infrastructure comparison and currently costs 2 credits per browser minute.

This is why a raw “credits versus credits” comparison is misleading. The credit has a different effective dollar value depending on the Firecrawl plan.

Using the included annual-billing price and monthly credit pool as a rough nominal value:

• On Hobby, one included credit is about $0.0032, so 2-credit Browser time is roughly $0.0064/minute.
• On Standard, one included credit is about $0.00083, putting the same 2-credit Browser minute at roughly $0.00166/minute.
• In comparison, a TinyFish Browser costs $0.002 per minute.

So TinyFish is cheaper than the nominal Hobby example but more expensive than Standard. Compare the workflow in dollars on the tier you would actually use.

The products differ too: Firecrawl Interact layers actions onto a scrape flow, while TinyFish Browser is managed browser infrastructure your Playwright or CDP code drives. Compare standalone Browser products when browser infrastructure is the job and Agent products when the job is autonomous execution.

TinyFish’s direct read path removes another meter entirely. TinyFish Search costs $0 per request, with current rate limits of 30 requests per minute with increased rate limits available ala carte. The Fetch API also costs $0 per URL, accepts up to 10 URLs in one request, and currently offers 150 URLs per minute with increased rate limits available.

Worth knowing: Search-heavy and recurring jobs are where endpoint pricing compounds fastest. With Firecrawl, Search can stack with Scrape, Monitor bills each repeated check, and browser interaction has a time-based meter. Model the entire sequence of calls rather than pricing the first endpoint in isolation.

Cost Line 3: Advanced Output and Agent, Where the 1-Credit Headline Breaks

A plain Firecrawl Scrape costs 1 credit per page. Structured extraction and other advanced options can push the same page well above that baseline.

For example, JSON-mode extraction costs 5 credits per page: 1 for the base Scrape plus 4 for JSON mode. PDF parsing adds another credit per PDF page. Firecrawl also documents additional 4-credit charges for options such as Question, Highlights, enhanced proxy, and PII redaction when they are used.

So if your workflow depends on structured JSON rather than plain Markdown, “pages included” can be a misleading way to estimate capacity. A 100,000-credit Standard plan represents up to 100,000 basic 1-credit pages, but only 20,000 pages if every request uses 5-credit JSON extraction.

Firecrawl Agent introduces a different meter. It is currently in Research Preview with five free runs per day, followed by dynamic credit-based pricing that varies with the complexity of the task. Firecrawl lets you set maxCredits to cap a run, so spend is bounded, but the exact cost is still not known until the work is done.

That said, the API bill is only one side of extraction economics. The other is how much text your model has to process after the page comes back.

In our June 2026 Fetch-quality evaluation, we ran five providers across 45 real URLs using their default configurations. TinyFish returned usable content on 42 of 45 URLs, or 93%, with a 90.5% median signal ratio. Firecrawl returned usable content on 28 of 45 URLs, or 62%, with a 79.5% median signal ratio.

The difference was not uniform. Firecrawl matched TinyFish 6/6 on pricing pages, while most of the coverage gap appeared on news sites. No provider won every category.

One Daily Mail article shows why output cleanliness can affect cost downstream. TinyFish returned 4,737 characters at 86.7% signal. Firecrawl returned 171,095 characters at 19.3% signal. For that specific page, we estimated an approximately 36x difference in downstream token cost because the model had to process far more surrounding navigation, ads, and related content to reach roughly the same article body.

This is a single tested example, not a universal 36x claim. The evaluation used 45 URLs, TinyFish ran the test, and web products change quickly. But the underlying pricing lesson holds: the amount of usable context returned per fetch affects the LLM bill after the scraping bill.

Our Fetch API also gives you a few controls for recurring jobs. It can return Markdown, HTML, or a structured JSON tree, process up to 10 URLs in one request, and isolate failures per URL so one bad page does not fail the whole batch. You can use ttl: 0 when you need a live pull, or accept a cached result within a freshness window when you do not.

For pages you revisit, conditional requests add another lever. Save the returned ETag or Last-Modified value, replay it on the next request, and an unchanged page can return not_modified: true so your application skips reprocessing the same content.

PDF text extraction is supported too, while binary files such as images and video return an error rather than pretending there is useful text to process.

Firecrawl has also shipped formats designed to reduce that downstream context, so the right test is your own page mix rather than assuming one provider wins every URL. For a deeper look at the downstream cost side of web extraction, read why 80% of a web fetch can be junk.

Pro Tip: When you estimate extraction cost, model credits per useful result plus downstream tokens, not only credits per URL. The cheapest API call is not necessarily the cheapest input for the model that reads it.

Cost Line 4: Firecrawl via MCP and CLI is an Access Surface, Not a New Meter

Using Firecrawl through MCP or its CLI changes how you call the product, not the underlying pricing.

The Firecrawl MCP server exposes Scrape, Crawl, Search, extraction, and Agent capabilities to MCP-compatible clients, while its CLI maps terminal commands back to the same API operations. Search still follows the Search meter, a 5-credit JSON Scrape still costs 5 credits, and Agent still follows dynamic Agent pricing.

We expose TinyFish through MCP as well, with quick-install support for Claude Code, Claude Desktop, Cursor, Codex, ChatGPT, Grok, Windsurf, and other MCP-compatible clients.

The current MCP economics match our direct product meters:

So Search and Fetch remain free whether you call them directly or through MCP. Agent and Browser still use the same step-based and time-based credit meters.

Worth knowing: MCP and CLI are access surfaces, not separate pricing products. Price the underlying web operation.

Where Firecrawl Costs Add Up

Firecrawl publishes its main credit rates clearly. The harder part is estimating how those credits behave once the workload moves beyond plain, steady scraping.

• Unused credits do not roll over on most self-serve plans. If your workload is quiet for a month, unused credits expire instead of carrying forward. Firecrawl currently offers rollover on Scale and Enterprise, but Hobby, Standard, and Growth users need to size recurring capacity around the billing cycle.
• Target-site errors can still be billable. Firecrawl does not charge when the failure happens on Firecrawl’s side, such as a timeout or server error. If Firecrawl successfully reaches the target and that site returns a 4xx or 5xx response, the request is still charged.
• Advanced output costs more than the base page. JSON mode is 5 credits per page, while other options such as enhanced proxying, Question, Highlights, PII redaction, and PDF parsing can add credits to the request. A plan’s advertised page equivalent only holds when you stay on the base 1-credit path.
• There is no general pay-as-you-go plan beyond Free. Once 1,000 free monthly credits are not enough, you move onto a recurring plan. That can work well for steady crawling and extraction, but it is less convenient for teams whose usage arrives in bursts.
• Agent pricing is variable. Firecrawl gives five Agent runs per day free during Research Preview, then charges dynamically according to the work performed. maxCredits lets you set a ceiling, but there is no fixed per-run rate to put into a budget beforehand.

There is some community evidence behind the budgeting concern. In one developer discussion on Reddit about Firecrawl, users asked why individual site scrapes sometimes appeared to consume multiple credits.

Firecrawl’s current documentation now makes at least part of that behavior easier to explain: one URL does not always equal one credit once advanced features are switched on.

Firecrawl Pros & Cons

Firecrawl has earned a strong developer following for a reason. The question is less whether the product works and more whether its pricing shape matches the workload you plan to run.

Pros

• Strong crawl-to-Markdown experience. Firecrawl is built around getting web content into clean Markdown and other model-ready formats without making developers maintain their own rendering and extraction infrastructure.
• A practical free tier. The 1,000-credit monthly allowance gives you enough room to test basic scraping and crawling before deciding whether the hosted product fits your pipeline.
• Public credit pricing. Scrape, Crawl, Map, Search, Monitor, and Interact all have published base meters. Even Agent, while dynamic, exposes credit use and supports a maximum-credit ceiling.
• Wide developer ecosystem. Firecrawl supports SDKs, MCP, CLI workflows, integrations, and has open-source roots, giving developers several ways to fit it into an existing stack.
• Predictable for straightforward scrape and crawl jobs. If you mainly turn known pages into Markdown, run at a steady volume, and rarely switch on higher-cost options, it is relatively easy to estimate consumption.

Cons

• Credit usage gets harder to predict once advanced features enter the job. A one-credit Scrape becomes five credits with JSON mode, while AI interaction, enhanced access, document parsing, and recurring monitoring introduce their own meters.
• Unused self-serve credits can disappear between busy periods. A Firecrawl user running bursty n8n research agents described the product itself as solid but found the monthly subscription awkward because unused credits vanished during quiet periods.
• A billable request does not always mean useful content came back. When the target website itself returns an HTTP error after Firecrawl reaches it successfully, the request can still consume credits.
• Agent is still in Research Preview. It can autonomously search and gather data, and you can cap its spend with maxCredits, but dynamic pricing gives production teams less certainty than a fixed per-step or per-run meter.
• The base page allowance can understate real consumption. If your application routinely needs structured extraction rather than plain page content, a plan that appears to cover 100,000 basic pages can support far fewer completed structured extractions.

Want the engineering view behind the page-read side of the bill? See how production-grade web fetching works for AI agents, including rendering, extraction, batch isolation, and downstream context quality.

And if your workload moves between reading pages and operating them, run the same job through TinyFish before you commit to a larger credit bundle. Compare the completed workflow and its total cost, not one API request in isolation.

Who Is Firecrawl Best For?

Firecrawl makes the most sense when the workload lines up with the parts of its pricing model that are easiest to forecast.

Developers Building Crawl-to-Markdown Pipelines

Firecrawl is a strong fit for developers who know the sites they need to crawl and primarily want page content in clean Markdown.

A steady documentation-ingestion, RAG, content-indexing, or site-crawling pipeline stays close to the 1-credit-per-page base rate. Higher plans also increase concurrency significantly, which gives high-volume jobs more room to fan out.

If your feature use stays consistent, the credit model is straightforward enough to budget around.

Prototypers and Early-Stage Builders

The 1,000-credit free plan gives early-stage teams room to find out whether Firecrawl solves their extraction problem before paying.

That works well for proofs of concept, small research utilities, internal prototypes, and products that have not yet established their production page volume.

Firecrawl also gives builders a broad set of surfaces to experiment with, so the free tier can help you work out whether your eventual workload is primarily Scrape, Search, Agent, or browser interaction.

Teams Running Recurring, Multi-Step, or Authenticated Workflows

Once a job grows from reading pages into operating websites, the cost model becomes more layered.

Firecrawl is not limited to scraping. It now offers Agent, browser infrastructure, Interact, and persisted browser-profile capabilities. Teams evaluating it for multi-step work should test those products directly rather than assuming the platform stops at extraction.

But this is also where we recommend comparing TinyFish closely. Search and Fetch remain free in our platform, while incremental costs begin when your workflow moves into Agent or Browser work. For authorized authenticated jobs, our Agent can also run with saved session state and credential handling when the workflow needs to get through a sign-in and complete work behind it.

Firecrawl Customer Reviews

Firecrawl has an active developer community, so Reddit gives a useful overview of what people like in day-to-day workflows and where the pricing model creates friction.

Positives

The recurring positive theme is convenience.

In a May 2026 Firecrawl discussion, a developer building an AI data pipeline preferred Firecrawl as the hosted option because it was more reliable in their setup and returned clean Markdown that could go straight into an LLM. They contrasted that with the extra infrastructure and debugging work involved in running a self-hosted alternative.

Complaints

Most of the pricing complaints are about how usage fits the billing cycle, rather than whether Firecrawl should charge at all.

The n8n user above had bursty research workflows, with heavy execution during one week and little activity the next month. Their frustration was paying a recurring subscription while unused credits expired.

In the Reddit discussion that we touched upon earlier, several users raised pricing as a concern, including questions about why a single website scrape could sometimes use multiple credits. A more recent Hermes Agent discussion shows the same issue from a lighter-use workflow. One commenter said a research session exhausted their Firecrawl credit supply, while another said Firecrawl had been too expensive during testing.

Those experiences do not mean Firecrawl is expensive for every workload. The recurring complaint is narrower: bursty usage and workflows that activate several meters make the plan-card price a weaker predictor of the eventual bill.

Alternative to Firecrawl: TinyFish

Pricing is only half the fit question. Firecrawl has moved further into interactive and agentic workflows too. Its Agent accepts natural-language tasks and structured-output schemas, while Interact can run prompts or Playwright code against a browser session and persist state across sessions.

So we would not frame this as “Firecrawl scrapes, TinyFish automates.” The more sensible question is what kind of web job sits at the center of your application.

TinyFish gives developers Search, Fetch, Browser, and Web Agent on one platform. You can start with the read path, then move into operation without stitching separate search, extraction, browser-state, and execution systems together. We explain the engineering trade-off in Why Stitched Web Stacks Fail in Production.

The TinyFish pricing table above covers the meter. Here, the difference is what happens when the job moves from retrieval into operation.

Goal-Based Web Agent

With the TinyFish Agent API, you pass a URL and a plain-English goal. Web Agent can navigate the site, work through filters and pagination, fill forms, authenticate, and return the result.

When the downstream system needs a predictable shape, pass output_schema. TinyFish validates the schema before execution and returns the final run result in the requested JSON shape.

You can run the same goal three ways:

Firecrawl Agent also accepts natural-language prompts, optional URLs, JSON schemas, and a maxCredits ceiling. Interact can drive browser actions through prompts or code. That makes Firecrawl a legitimate agentic option, not simply an extraction API.

Our differentiated fit is recurring web operation where the application needs the read path, browser execution, session state, and authenticated work to function together.

Authenticated Workflows

For authorized logged-in work, Browser Context Profiles persist cookies, local storage, and session storage between Agent runs.

Pair the profile with Vault credentials when the workflow needs to recover from an expired login. The profile carries the existing session forward; if that session becomes stale and a matching Vault credential is available, TinyFish can use it to repair the session.

Firecrawl also supports persistent browser profiles, so persistence alone is not the differentiator. TinyFish combines persistent Agent state with Vault-backed credential handling for recurring authorized workflows.

Normal site controls still apply. Sessions expire, some OAuth or SSO flows may not work, and CAPTCHAs or bot challenges can still stop a run.

For example, an operations team could keep an authorized distributor session in a profile, ask Web Agent to open the inventory dashboard, filter low-stock products, work through every result page, and return the rows in a defined JSON schema. /run-async lets the application start that longer workflow and collect the result when it finishes.

Have a workflow that keeps growing beyond “fetch this page”? Run it in the TinyFish Playground against the real site and see how much of the orchestration you can remove.

Does Firecrawl Pricing Fit Your Budget?

Firecrawl pricing fits best when your workload is predictable enough to map cleanly to its credit system.

If you primarily crawl public websites and turn pages into Markdown, its 1-credit base path is easy to understand, and the free tier gives you a fair way to test the workflow before committing. Higher-volume plans can also drive the nominal cost per basic page down considerably.

But if your application spends more time searching for pages, repeatedly fetching them, operating websites, or running authenticated multi-step workflows, model the whole job before choosing a plan. The headline page allowance becomes much less useful once Search, advanced outputs, Interact, monitoring, or Agent work enter the sequence.

That is one of the areas where we built TinyFish differently. Search and Fetch give your agent the web-reading layer, while Browser and Web Agent take over when the job requires direct control or autonomous operation.

You can start free with TinyFish, or read the full head-to-head comparison.

And whichever platform you choose, do not optimize the web layer in isolation from the model consuming its output. Give agents better context, not more context, as cleaner retrieval reduces the amount of context your model has to process after the API call.

The final decision comes down to this: price the workflow you expect to run in production, not the cheapest-looking primitive on the pricing page.

FAQs

1. How much does Firecrawl cost?

Firecrawl currently has a Free plan at $0, followed by Hobby at $16/month, Standard at $83/month, Growth at $333/month, and Scale at $599/month when billed yearly. Enterprise pricing is custom.

2. Can I use Firecrawl for free?

Yes. Firecrawl’s Free plan includes 1,000 credits per month and 2 concurrent requests. Firecrawl Agent also currently includes five free runs per day during Research Preview.

3. How much does scraping cost in Firecrawl?

A basic Firecrawl Scrape starts at 1 credit per page. Crawl and Map also start at 1 credit per page, while advanced output and processing options can increase the effective cost.

4. Do Firecrawl credits roll over?

Not on most self-serve plans. Unused credits generally expire at the end of the billing period rather than rolling into the next one.

5. What is a cheaper alternative to Firecrawl for Search and Fetch?

TinyFish charges 0 credits for direct Search requests and Fetch URLs. Agent and Browser are metered separately, so compare the complete workflow rather than one endpoint alone.