CAPTCHA Solving Service

A CAPTCHA solving service accepts a challenge from your crawler and returns something you can submit. Two delivery models exist. Token services handle reCAPTCHA v2 and v3, hCaptcha and Cloudflare Turnstile: you send the site key and page URL, they return a solved response token, and you inject it into the corresponding form field before submitting. Image and click services return a transcription or a set of coordinates for the older visual puzzles, and your code performs the interaction itself. The token model is what most scraping stacks use now.

Pricing is quoted per thousand solves and spreads widely by challenge type. 2Captcha's public pricing page on 2026-07-27 listed normal captcha at €0.5 to €1 per 1,000, reCAPTCHA v2 at €0.99 to €2.8 per 1,000, reCAPTCHA v3 at €1.4 per 1,000 for score 0.3 and below and €2.8 above it, and Cloudflare Turnstile at €1.4 per 1,000. Those figures move, so treat the shape rather than the number: the harder the challenge, the higher the rate, and score-based challenges are priced by the score you ask for. Latency is the cost that actually hurts and it is not something we have measured. Reported solve times run into seconds and tens of seconds depending on type and queue depth, which is several orders of magnitude above your normal request time.

Volume in this category is falling for a structural reason. Modern defenses score the session before a challenge is ever shown. Cloudflare describes its managed challenge as dynamically choosing what to serve "based on the characteristics of a request," with most human visitors verified without interacting at all (Cloudflare challenge pages documentation, read 2026-07-27). A token bought for a session that already looks automated can be perfectly valid and still fail, and reCAPTCHA v3 returns a score rather than a verdict, so the site decides what to do with it. Buying solves does not repair a session; it pays for the symptom.

The budget question follows from that. The same money spent on session quality upstream, meaning better IP sourcing, a coherent fingerprint and human-plausible pacing, stops challenges firing in the first place. Managed vendors such as Scrapfly, ZenRows, Scrappey and Bright Data fold both into a per-request price, which hides the split from you. Keep dedicated solving capacity for the tail of targets that challenge every session regardless. Then size concurrency around solve latency rather than price: if a solve occupies a worker for thirty seconds, that worker's throughput collapses, and you need proportionally more workers for the same page rate.