ScrapeGraphAI (scrapegraphai.com) sells two things under one name. Its free Python library, under the MIT licence, turns a plain-language prompt into a scraper using a language model you supply. Its hosted API does the same for credits. On 22 April 2026 it released V2, a rewrite of the API: the same monthly prices now buy two to three times the credits, and an AI extraction costs 5 credits, down from 10. Searches for the name went from 3,600 a month to 9,900 in May and June. Below are its prices before and after, what the library leaves to you, what five other pages say and the alternatives.
The library and the API
The library came first. Its GitHub repository dates from 27 January 2024 and its first release on PyPI from 17 February 2024. By 13 October 2026 it counted 490 releases, the latest on 6 October, 31,733 stars on GitHub and 32,537 downloads in the past month. It needs Python 3.12 or newer and a language model: OpenAI, Groq, Gemini or Azure through their APIs, or a local model through Ollama.
Its own README lists the browser setup, proxies, anti-bot work and scaling as your job. The browser loader takes a proxy server with a username and password, and can also pick addresses from public free proxy lists through the free-proxy package.
The API belongs to ScrapeGraphAI, Inc. Its V2 has five actions, scrape, extract, search, crawl and monitor, with SDKs for Python and JavaScript under MIT. The price page says every request is "SOC 2 Type II compliant".
What the API costs
ScrapeGraphAI's API plans, 13 October 2026
| Plan | A month | Credits a month |
|---|---|---|
| Free | $0 | 500, once |
| Starter | $20, or $17 paid yearly | 10,000 |
| Growth | $100, or $85 paid yearly | 100,000 |
| Pro | $500, or $425 paid yearly | 750,000 |
scrapegraphai.com/pricing and /compare, read from our server on 13 October 2026
Rate limits run from 10 requests a minute on the free plan to 5,000 on Pro, and proxy rotation appears from Growth, basic there and advanced on Pro. A page costs 1 credit as markdown, 2 as a screenshot and 25 as a branding analysis. An AI extraction costs 5, a search 2 per result or 5 with a prompt, a PDF 1 per page, and a crawl adds 2 to start.
The stealth setting, an anti-bot bypass, adds 5 credits to a scrape or an extraction and to each search result, and in auto mode a simple markdown page "could be as high as 6 credits". Packs of extra credits cost $5 for 1,000, $40 for 10,000 and $150 for 50,000 and never expire. Plan credits do not carry over between billing cycles and, under the terms of 4 June 2026, are non-refundable.
Before and after V2
From at least February 2025 to April 2026 the same $20, $100 and $500 bought 5,000, 40,000 and 250,000 credits. The AI extractor, then called SmartScraper, cost 10 credits a page, Markdownify 2 and a search 30. V2 doubled the Starter credits, multiplied Growth's by two and a half and Pro's by three, and halved the price of an extraction. On the $100 plan an AI extraction fell from 2.5 cents to half a cent.
Who runs it
The terms of 4 June 2026 name ScrapeGraphAI, Inc. and place disputes under "the laws of the jurisdiction where ScrapeGraphAI, Inc. is established", without naming that jurisdiction. OpenCorporates lists SCRAPEGRAPHAI, INC. in Delaware, with a branch registered in California at a San Francisco address. The privacy policy, last updated on 15 March 2024, says personal data goes to the United States. The domain was registered on 29 April 2024. Liability under the terms is capped at three months of fees or $100, whichever is greater.
What other pages say about ScrapeGraphAI
- Firecrawl's comparison, built on a benchmark it ran on 13 January 2026, says it "delivers roughly 5x more extracted pages" per dollar, at "
$0.021/page" for ScrapeGraphAI against "$0.004/page". Those were the old prices; at V2 prices the two are level. It also says ScrapeGraphAI "has achieved SOC 2 Type 1 and is working toward Type 2", while ScrapeGraphAI's price page now claims Type II for every request. - Scrapfly's page, titled "ScrapeGraphAI alternative that actually works?", says teams leave for features it lacks, "JS rendering, Extraction API, unified billing", and that extraction "Requires SmartScraper (10 credits)". V2 sells a JavaScript render mode and a stealth bypass, and an extraction costs 5 credits.
- Bright Data's tutorial (October 2024, updated September 2025) runs the library with OpenAI's GPT-4. It says nothing of what the model's tokens cost or of the hosted API.
- Decodo's overview (30 April 2026, updated 24 July 2026) explains how the library cleans HTML and splits it into chunks that fit a model's context window. The hosted API's prices are missing.
- ScrapingBee's tutorial runs the library on a local LLaMA model through Ollama and warns that bigger models cost more in hardware and electricity.
The alternatives
Alternatives to ScrapeGraphAI, October 2026
| Tool | What it is | To start |
|---|---|---|
| Firecrawl | Hosted API; open source under AGPL-3.0 | 1,000 free credits a month; $19 for 5,000 |
| Crawl4AI | Open-source library, Apache-2.0 | Free; your own model and proxies |
| Scrapfly | Hosted API with anti-bot and extraction | 1,000 free credits; $30 for 200,000 |
| ScrapeGraphAI library | Open-source library, MIT | Free; your own model and proxies |
firecrawl.dev and scrapfly.io price pages, GitHub and PyPI, read from our server on 13 October 2026
Firecrawl returns a page as markdown for 1 credit and charges 5 for AI extraction. A scrape that returns nothing is free, an error page costs 1 credit, and Hobby, Standard and Growth credits lapse at the end of the month. Its repository counts 190,855 stars on GitHub. Crawl4AI, with 85,388 stars and 1.28 million downloads in the past month, turns pages into markdown for language models and leaves the model and the proxies to you. Scrapfly prices by credits that vary with how hard the target is; its page claims 98% success on blocked sites, which we did not test.
Where HProxy fits
Run the library or Crawl4AI yourself and the proxies are yours to bring. Our residential proxies start at $0.44 per GB for rotating addresses, and Premium adds city and ISP targeting with sticky sessions of up to 24 hours from $1.00 per GB. Datacenter addresses cost $1.58 per IP a month. ScrapeGraphAI's loader takes such a proxy as a server, a username and a password.
Which one to pick
- ScrapeGraphAI's API for prompts in and JSON out from $20 a month, with stealth only on the sites that need it, since it doubles the cost of an extraction.
- Firecrawl for the same price per extraction, markdown at 1 credit a page and a larger open-source project behind it.
- Crawl4AI or ScrapeGraphAI's library to run everything yourself, paying only for the model and the proxies.
- Scrapfly if blocked sites are the problem, after a test on your own targets.
What we could not find out
We ran no scrapes, so success rates, speed and the quality of the extracted data are unknown. So is what the stealth setting does on a given site. What the library costs depends on the model and on page size, which we did not measure. We did not work out Scrapfly's credits per extraction. ScrapeGraphAI's price pages and the libraries' records get another reading by 13 January 2027.
Sources
- scrapegraphai.com: the price page, the V2 announcement of 22 April 2026, the comparison page, the terms of 4 June 2026 and the privacy policy, read from our server on 13 October 2026.
- The Internet Archive's copies of scrapegraphai.com/pricing, February 2025 to September 2026.
- GitHub (ScrapeGraphAI's repositories, the library's README and its loader code) and PyPI with pypistats, 13 October 2026.
- OpenCorporates and Verisign's registry record for scrapegraphai.com, 13 October 2026.
- Pages about ScrapeGraphAI by Firecrawl, Scrapfly, Bright Data, Decodo and ScrapingBee, read 13 October 2026.
- The price pages of Firecrawl and Scrapfly, and the GitHub and PyPI records of Crawl4AI and Firecrawl, 13 October 2026.
- Search counts for ScrapeGraphAI, Firecrawl and Crawl4AI from Google Keyword Planner, worldwide, 13 October 2026.
- Our own residential and datacenter prices, filled in from live price data when the page is built.
