Every scraper developer shares a rite of passage: build something beautiful, watch it hum along on 10 test pages, and then—wall of CAPTCHAs. Pipeline dead.
I've lived this cycle more times than I care to admit, both from my years in automation at Automation Anywhere and now building Thunderbit. The thing is, CAPTCHAs in 2026 aren't the squiggly-text puzzles of 2010. Modern challenges like reCAPTCHA v3, Cloudflare Turnstile, and hCaptcha analyze your browser fingerprint, mouse behavior, TLS handshake, and session history before you even see a checkbox.
According to the HTTP Archive's 2025 Web Almanac, reCAPTCHA and Cloudflare Bot Management collectively protect about 30% of measured websites. And with bad bot traffic hitting 37% of all internet traffic in 2024 (up from 32% the year before), sites are only getting more aggressive. This guide covers 8 CAPTCHA solvers I've researched and tested at scale—but it also introduces a prevention-first framework that, for many use cases, eliminates the need for a solver entirely.
Before You Solve, Try Prevention: The CAPTCHA Avoidance-First Framework
After years in the scraping and automation world, one lesson keeps proving itself: the cheapest CAPTCHA to solve is the one you never trigger. Most CAPTCHA encounters during web scraping aren't inevitable—they're symptoms. The site's risk engine didn't like your IP, your request rate, your browser fingerprint, or your session history. Fix those, and the CAPTCHA often disappears.
Before reaching for a CAPTCHA-solving API, walk through this decision tree. I think of it as Step Zero.
The CAPTCHA Prevention Decision Tree
- Does the site offer a public API, data feed, or export? → Use it. No CAPTCHA, no scraping complexity.
- Is the target a popular platform (Amazon, Zillow, LinkedIn, Shopify, Instagram)? → Try pre-built scraper templates first. Thunderbit, for example, offers instant templates for these sites that handle anti-bot measures in the cloud—no CAPTCHA interaction needed.
- Are CAPTCHAs only appearing when you scale up? → Fix your request hygiene: throttle requests, rotate residential proxies, persist sessions, align geo/timezone/language, and ensure your browser/TLS fingerprints look coherent.
- Are CAPTCHAs appearing immediately from a cloud or server runtime? → Compare against a real browser session. Datacenter IPs, containers, and default headless browsers are flagged more aggressively.
- Still hitting CAPTCHAs after all prevention steps? → Now choose a solver from the list below.
Common Avoidance Strategies That Actually Work
| Strategy | Why It Works | Practical Guidance |
|---|---|---|
| Request throttling & rate limiting | Fixed-interval fast requests scream "bot" | Randomize timing, model pages/minute, monitor 403/429/CAPTCHA rates |
| Residential/mobile/ISP proxies | Datacenter IPs have lower reputation on consumer sites | Use sticky sessions when tokens/cookies bind to IP |
| Browser fingerprint management | Sites compare user-agent, viewport, canvas, WebGL, fonts, timezone, language | Keep attributes consistent; align Accept-Language, timezone, and proxy country |
| TLS/HTTP/2 fingerprint consistency | Cloudflare uses JA3/JA4 to fingerprint TLS connections | Use real browsers or tools that mimic browser TLS behavior; raw Python requests often looks non-browser |
| Session persistence & cookies | Risk scores improve with coherent browsing history | Reuse cookies/localStorage for a session; rotate only when blocked |
| Headless stealth configuration | Default headless signals get flagged | Use Puppeteer/Playwright stealth plugins, but verify against real target behavior |
A 2024 arXiv study of evasive bot traffic found that adding fingerprint-consistency rules reduced bot evasion by 48% against DataDome and 45% against BotD across 500,000 requests. That's a massive signal: fingerprint coherence matters more than most people think.
Thunderbit's cloud scraping infrastructure handles all of this behind the scenes—rotating IPs, managing fingerprints, handling anti-bot measures—so business users in sales, ecommerce, and real estate never encounter CAPTCHAs at all. My team built it specifically so you pay per output row, not per CAPTCHA challenge. More on that below.
Try Thunderbit for CAPTCHA-Free Scraping
Understanding CAPTCHA Types: What Your Scraper Is Up Against
The CAPTCHA landscape in 2026 is a zoo, and not every solver handles every animal. Knowing what you're up against matters before you pick a tool.
reCAPTCHA v2 (Checkbox and Image Challenges)
Google's familiar "I'm not a robot" checkbox. It analyzes browsing history, mouse movements, and browser behavior. Suspicious users get image selection challenges (traffic lights, crosswalks, fire hydrants—you know the drill). Google's reCAPTCHA v2 docs describe the widget rendering and customization options.
Difficulty: Medium. A USENIX Security 2025 benchmark found generalist visual solvers cracking reCAPTCHA v2 at 68-72%. Most commercial solvers handle it well.
reCAPTCHA v3 (Invisible Score-Based)
Runs invisibly in the background. Returns a score from 0.0 (likely bot) to 1.0 (likely human). The website decides the threshold—and that's the catch. Google's v3 docs recommend starting at a 0.5 threshold. Tokens expire after two minutes. Google has moved reCAPTCHA under Google Cloud Fraud Defense, adding enterprise risk reasons like AUTOMATION, UNEXPECTED_ENVIRONMENT, and TOO_MUCH_TRAFFIC.
Difficulty: Hard. A solver can return a v3 token, but whether it passes depends on the site's threshold, your browser state, IP reputation, and traffic pattern.
hCaptcha
A privacy-focused alternative to reCAPTCHA. Similar image-selection tasks, but Enterprise supports invisible and passive modes. hCaptcha's privacy policy says it collects mouse movements, scroll position, keypress events, and touch events.
Difficulty: Medium to High. USENIX 2025 found hCaptcha visual challenges solved at 82% by the best solver, but Enterprise/passive modes add risk scoring.
Cloudflare Turnstile
Cloudflare calls Turnstile a "smart CAPTCHA alternative" that can work without showing visitors a visual challenge. It runs non-interactive JavaScript challenges gathering browser signals—proof-of-work, proof-of-space, web API probing, browser-quirk detection, and human-behavior signals. With pre-clearance, a Turnstile widget can issue a cf_clearance cookie to bypass later WAF challenges on the same zone.
Difficulty: Hard to Very Hard. The challenge isn't an image puzzle—it's maintaining coherent TLS/browser/IP/session context.
FunCaptcha (Arkose Labs), GeeTest, and Others
Arkose Labs uses interactive game-like puzzles: rotate objects, match dice, drag shapes. Third-party comparisons claim 1,250+ puzzle variations. USENIX 2025 found wide variation in AI success: Arkose 3D rollball 20%, dice match 71%.
GeeTest v4 includes Slide Puzzle, Icon Selection, Match-Three, Gobang, and invisible probe modes. AWS WAF has its own CAPTCHA and Challenge actions. Friendly Captcha uses proof-of-work instead of visual puzzles.
Simple image/text CAPTCHAs are Low difficulty in 2026—commodity solvers handle them easily.
How We Evaluated the Best CAPTCHA Solvers for Web Scraping
I evaluated every tool against a standardized framework to keep this comparison apples-to-apples:
- Supported CAPTCHA types — breadth of coverage
- Solving method — AI, human workers, hybrid, or avoidance-based
- Average solve speed under real-world conditions
- Pricing per 1,000 solves and cost at scale
- API and language support — Python, JS, PHP, Go, etc.
- Proxy integration and compatibility
- Free tier or trial availability
- Concurrency handling and rate limits
- Best-fit use case — who should actually use this tool
I cross-referenced official documentation, current pricing pages, community feedback on Reddit and Stack Overflow, and independent benchmarks like the USENIX Security 2025 CAPTCHA study. Where pricing is dynamic, I used the rates posted at the time of research.
Best CAPTCHA Solvers for Web Scraping at a Glance
The master comparison table is below. Scan it first, then dig into the individual reviews.
| Tool | Solving Method | reCAPTCHA v2/v3 | hCaptcha | Cloudflare Turnstile | FunCaptcha | Price per 1K Solves | Avg Speed | Free Tier | Best For |
|---|---|---|---|---|---|---|---|---|---|
| Thunderbit | AI avoidance / managed scraping | N/A (bypassed) | N/A (bypassed) | N/A (bypassed) | N/A | Credit-based (~1 credit/row) | Cloud: up to 50 pages at once | ✅ Free 6 pages/mo | Non-devs who want data, not solver plumbing |
| CapSolver | AI-first | ✅ / ✅ | Verify current docs | ✅ | Not in current pricing | v2 $0.80, v3 $3.00, Turnstile $1.20 | Often 1-10s | Trial varies | Fast Cloudflare/JS-heavy scraping |
| 2Captcha | Human + automation | ✅ / ✅ | ✅ | ✅ | ✅ | v2 $1-2.99, Turnstile $1.45 | ~13s reCAPTCHA | Small trial | Broadest CAPTCHA type coverage |
| Anti-Captcha | Human-worker network | ✅ / ✅ | ✅ (in API docs) | ✅ | ✅ | v2 $0.95-1, Turnstile $1.00 | ~5-10s | Deposit-based | Accuracy and queue transparency |
| CapMonster Cloud | AI-first | ✅ / ✅ | Verify current docs | ✅ | ✅ | v2 $0.60, Turnstile $0.30 | 1-5s common | $0.10 test balance | Drop-in 2Captcha/Anti-Captcha replacement |
| AZcaptcha | AI/OCR | ✅ / ✅ | Not officially listed | ✅ | Not officially listed | v2 $1.00, Turnstile $0.90 | v2 8-15s, Turnstile 5-10s | $0.10 free credit | Budget common CAPTCHA solving |
| EndCaptcha | Human with SLA | ✅ / Claimed | Not officially listed | Not officially listed | Not officially listed | Starter $11.99/3K credits | ~7-8s (70%), ≤11s | 1,000 free credits | SLA-backed image/text workflows |
| Best Captcha Solver | 100% human | ✅ / ✅ | Not officially listed | ✅ | ✅ | reCAPTCHA $1/1K, Turnstile $0.80/1K | ~14s regular, ~46s reCAPTCHA | Demo/account | Human accuracy for complex formats |
A note on Thunderbit: Unlike the other seven tools, Thunderbit doesn't sell CAPTCHA tokens. It avoids triggering challenges in the first place through managed cloud scraping. You pay per output row, not per CAPTCHA. For many business users, this means CAPTCHA-solving costs go to zero.
1. Thunderbit — Skip the CAPTCHA Entirely with AI-Powered Scraping
Thunderbit takes a fundamentally different approach from every other tool on this list. Instead of solving CAPTCHAs after they appear, Thunderbit's cloud scraping infrastructure avoids triggering them in the first place—handling JavaScript rendering, anti-bot protection, IP rotation, browser fingerprinting, and geo-routing behind the scenes. As someone who co-founded the company, I'm obviously biased, but the logic holds: if your scraper never encounters a CAPTCHA, you never need to pay to solve one.
The workflow is dead simple. Install the Thunderbit Chrome extension, navigate to any page, click "AI Suggest Fields," review the columns Thunderbit auto-detects, and hit "Scrape." The data lands in a structured table you can export to Excel, Google Sheets, Airtable, Notion, CSV, or JSON. No proxies to configure, no CAPTCHA tokens to inject, no queue logic to manage.
Key features:
- Cloud scraping processes up to 50 pages simultaneously, handling anti-bot measures in the cloud
- AI Suggest Fields auto-detects the data columns on any page—no manual CSS selector work
- Subpage scraping follows links to detail pages and enriches your dataset automatically
- Instant scraper templates for popular sites (Amazon, Zillow, Instagram, Shopify, and more)
- Free data export to Excel, Google Sheets, Airtable, Notion—no paywall on exports
- Credit-based pricing: 1 credit = 1 output row. Free tier includes 6 pages/month; Starter at $9/month with 5,000 credits/year; Pro at $16.50/month with 30,000 credits/year
For a sales team scraping leads from a Cloudflare-protected directory, or an ecommerce team monitoring competitor prices on Amazon, Thunderbit means you never see a CAPTCHA, never manage a solver API, and never worry about per-CAPTCHA costs ballooning at scale.
Best for: Non-technical business users—sales, ecommerce, real estate, ops—who want structured data without managing CAPTCHA-solving pipelines, proxies, or browser stealth configurations.
2. CapSolver — Best AI-Powered CAPTCHA Solver for Speed and Cloudflare
CapSolver is an AI-first CAPTCHA solving service built for speed. It uses machine learning models instead of human workers, which means faster solve times and no dependence on worker availability. Their official docs list support for reCAPTCHA v2/v3, reCAPTCHA Enterprise, Cloudflare Turnstile, Cloudflare Challenge, GeeTest, AWS WAF, DataDome, MTCaptcha, and ImageToText.
Key features:
- AI-driven solving with typical response times of 1-10 seconds for common types
- Strong Cloudflare Turnstile and Challenge support at $1.20/1K solves
- JSON REST API with community libraries for Python, Node.js, TypeScript, C#, PHP, and a Puppeteer plugin
- Proxy support including SOCKS4, SOCKS5, HTTP, HTTPS
- Chrome browser extension for non-developers
Pricing: reCAPTCHA v2 $0.80/1K, v3 $3.00/1K, Turnstile $1.20/1K, GeeTest $1.20/1K, AWS WAF $1.00/1K, ImageToText $0.40/1K.
Limitations: AI accuracy can temporarily drop when CAPTCHA providers update their detection logic. Current captured docs didn't expose a clear hCaptcha or Arkose pricing row—verify before committing if those are your primary targets.
Best for: Developers and technical teams scraping JavaScript-heavy or Cloudflare-protected sites at scale who prioritize speed and automation.
3. 2Captcha — Most Reliable Human-Powered CAPTCHA Solver
2Captcha is one of the oldest CAPTCHA-solving marketplaces, operating since 2007. It uses a global network of human workers, giving it the widest CAPTCHA type coverage I've found—their API docs list over 30 types, including reCAPTCHA v2/v3/Enterprise, Arkose/FunCaptcha, GeeTest v4, Turnstile, Capy, KeyCAPTCHA, Amazon, DataDome, Friendly Captcha, Tencent, and many more.
Key features:
- Broadest published CAPTCHA type support in the industry
- Dynamic pricing: reCAPTCHA v2 $1-2.99/1K, Turnstile $1.45/1K, Arkose $1.45-50/1K depending on difficulty
- Transparent capacity data: at capture time, free capacity showed 11,273/min for reCAPTCHA v2 and 12,000/min for image CAPTCHA
- Only charges for successful solves
- Official SDKs for Python, JavaScript/Node.js, Go, Ruby, C++, PHP, Java, and C#
Pricing: Average speed is ~13 seconds for reCAPTCHA, 5-15 seconds for simpler types, up to 30 seconds during peak hours. Their proxy docs note that proxies are not supported for reCAPTCHA v3/Enterprise v3 because they reduce success rate.
Limitations: Slower than AI-based solvers. Speed varies with worker availability. Costs add up at very high volumes.
Best for: Teams that need the broadest CAPTCHA type coverage and can tolerate slightly slower solve times—especially projects encountering unusual or custom CAPTCHA formats where AI models haven't been trained.
4. Anti-Captcha — Trusted Human Solver with Consistent Accuracy
Anti-Captcha is another long-running human-worker service (also since 2007). Its homepage claims 5-second solving speed, 99.99% uptime, and "unlimited parallel processing threads."
Key features:
- Supported types include reCAPTCHA v2/v3/Enterprise, FunCaptcha/Arkose, GeeTest, Turnstile, Amazon WAF, Friendly Captcha, Altcha, Prosopo, and custom AntiGate tasks
- Pricing: image CAPTCHAs $0.50-0.70/1K, reCAPTCHA v2 $0.95-1/1K, Turnstile $1.00/1K, Arkose $3/1K
- Queue transparency via getQueueStats API—you can check waiting workers, load percentage, speed, and total workers before sending tasks
- Proxy-based solving for IP-bound tokens
- SDKs for Python, JavaScript, PHP, C#, Go; browser extensions for Chrome, Firefox, Safari
Limitations: Speed varies with worker load. Complex CAPTCHAs can take 30-50 seconds. Slightly less breadth than 2Captcha's type list.
Best for: Users who value accuracy and queue transparency, and projects where CAPTCHA tokens must originate from a specific IP.
5. CapMonster Cloud — Fast AI Solver with Competitive Pricing
CapMonster Cloud is an AI-first solver from the CapMonster ecosystem. What makes it interesting for existing scraping setups: it's API-compatible with 2Captcha and Anti-Captcha, meaning you can often swap it in without changing your integration code.
Key features:
- Supported types: reCAPTCHA v2/v3/Enterprise, GeeTest v3/v4, Cloudflare Turnstile/Challenge, DataDome, Amazon WAF, FunCaptcha, Friendly Captcha, Tencent, Altcha, and more
- Pricing: reCAPTCHA v2 $0.60/1K (claimed 99% success), Turnstile $0.30/1K, FunCaptcha $1.10/1K
- Fast AI-based solving, typically 1-5 seconds for common types; FAQ says 1-30s depending on type
- SDKs for C#, Python, JS, Go, PHP
Limitations: Like other AI solvers, it can struggle with novel or rapidly changing challenge variants.
Best for: Teams already integrated with 2Captcha or Anti-Captcha who want faster speeds and lower per-solve costs without rewriting their code.
6. AZcaptcha — Budget-Friendly Solver with Low Per-Solve Rates
AZcaptcha positions itself as the most affordable option, especially for high-volume reCAPTCHA solving. Its pricing page says failed attempts are not charged.
Key features:
- Supported types: reCAPTCHA v2/v3, reCAPTCHA v2/v3 Enterprise, Cloudflare Turnstile, Image Captcha
- Pricing: reCAPTCHA v2 $1.00/1K, v3 $2.00/1K, Turnstile $0.90/1K, Image $0.40/1K. Bulk pricing above 10M/month cuts rates roughly in half
- Speed claims: reCAPTCHA v2 8-15s / 98% success, Turnstile 5-10s / 97%, Image 0.3-1s / 99%
- New accounts get $0.10 free credit after email verification; minimum top-up is $5
Limitations: Narrower type support than premium services. Less polished documentation. Accuracy claims may not hold at peak load.
Best for: Budget-conscious users with high-volume scraping projects focused on reCAPTCHA and Turnstile.
7. EndCaptcha — SLA-Backed Reliability for Mission-Critical Scraping
EndCaptcha differentiates with something rare in this space: a public Service Level Agreement. Their homepage advertises Slowness Insurance (approximately 70% of images decoded within 7-8 seconds, remainder guaranteed within 11 seconds) and Outage Insurance (credit compensation during qualifying outages, claims within 72 hours).
Key features:
- 1,000 free CAPTCHA credits to start
- Packages: Starter $11.99/3,000 credits, Big Dreams $43.99/10,000, Pro-Marketer $149.99/35,000, Guru $479.01/100,000
- SLA with uptime and speed guarantees—compensation if service expectations aren't met
- Primarily covers reCAPTCHA and image/text CAPTCHAs
Limitations: Current public pages are strongest for image/text and reCAPTCHA. I couldn't find clear support for hCaptcha, Turnstile, Arkose, GeeTest, or AWS WAF. The SLA excludes unsupported image types and maintenance windows.
Best for: Users whose scraping pipelines cannot tolerate downtime or slow solving—especially mission-critical ecommerce monitoring or time-sensitive data collection.
8. Best Captcha Solver — Human-Powered Accuracy for Complex Challenges
Best Captcha Solver (BCS) uses 100% human workers with its own typing team. It's been operating for 7+ years and supports migration from 2Captcha, DeathByCaptcha, and Anti-Captcha API formats.
Key features:
- Supported types: reCAPTCHA v2/v3, reCAPTCHA Enterprise, GeeTest v4, FunCaptcha/Arkose, Turnstile, image CAPTCHAs, Captcha Task
- Pricing: reCAPTCHA $1/1K, Enterprise $3/1K, Turnstile $0.80/1K, FunCaptcha $1.50/1K, Image $0.60/1K
- API libraries for multiple languages; no client-side browser required
- 24/7 human support and solving
Limitations: Slowest average speed on this list (~14 seconds for regular CAPTCHAs, ~46 seconds for reCAPTCHA). Not ideal for high-volume time-sensitive scraping.
Best for: Smaller-scale projects or those encountering unusual CAPTCHA formats where accuracy matters more than speed.
CAPTCHA Type × Solver Cross-Reference: Which Tool Handles What
This is the table I wish existed when I started researching CAPTCHA solvers—every reviewed tool mapped against every major CAPTCHA type, with difficulty ratings and indicative pricing.
| CAPTCHA Type | Difficulty | Thunderbit | CapSolver | 2Captcha | Anti-Captcha | CapMonster | AZcaptcha | EndCaptcha | BCS |
|---|---|---|---|---|---|---|---|---|---|
| reCAPTCHA v2 | ⚠️ Medium | N/A (avoided) | ✅ $0.80 | ✅ $1-2.99 | ✅ $0.95-1 | ✅ $0.60 | ✅ $1.00 | ✅ Credits | ✅ $1.00 |
| reCAPTCHA v3 | 🔴 Hard | N/A (avoided) | ✅ $3.00 | ✅ $1.45-2.99 | ✅ $2-3 | ✅ $0.90 | ✅ $2.00 | ⚠️ Claimed | ✅ Included |
| hCaptcha | ⚠️ Medium-High | N/A (avoided) | ⚠️ Verify docs | ✅ Listed | ⚠️ Verify docs | ⚠️ Verify docs | ❌ Not listed | ❌ Not listed | ❌ Not listed |
| Cloudflare Turnstile | 🔴 Hard | N/A (avoided) | ✅ $1.20 | ✅ $1.45 | ✅ $1.00 | ✅ $0.30 | ✅ $0.90 | ❌ Not listed | ✅ $0.80 |
| FunCaptcha/Arkose | 🔴 High | N/A | ⚠️ Not in pricing | ✅ $1.45-50 | ✅ $3.00 | ✅ $1.10 | ❌ Not listed | ❌ Not listed | ✅ $1.50 |
| GeeTest | ⚠️ Medium-High | N/A | ✅ $1.20 | ✅ $1.99 | ✅ $0.80 | ✅ $1.20 | ❌ Not listed | ❌ Not listed | ✅ $1-1.50 |
| AWS WAF | 🔴 High | N/A | ✅ $1.00 | ✅ $1.45 | ✅ $1.00 | ✅ $1.40 | ❌ Not listed | ❌ Not listed | ⚠️ Possible |
| Image/Text CAPTCHA | 🟢 Low | N/A | ✅ $0.40 | ✅ $0.50-1 | ✅ $0.50-0.70 | ✅ $0.30 | ✅ $0.40 | ✅ Credits | ✅ $0.60 |
Key takeaways: 2Captcha and Anti-Captcha offer the broadest human-worker coverage. CapSolver and CapMonster Cloud are the strongest speed/cost picks for common modern token types. AZcaptcha is budget-friendly but narrower. EndCaptcha's value is its SLA, not broad modern challenge coverage. And Thunderbit eliminates the entire question for users who need extracted data rather than a raw CAPTCHA token.
What 100K CAPTCHA Solves per Month Actually Costs
Per-solve pricing is one thing. What actually hits your budget at scale is a different conversation entirely. I've modeled monthly spend at three tiers using current public rates for Cloudflare Turnstile (one of the most common modern challenges).
| Service | 10K Solves/mo | 100K Solves/mo | 1M Solves/mo |
|---|---|---|---|
| Thunderbit | Credit-based—no per-CAPTCHA cost | Credit-based—no per-CAPTCHA cost | Credit-based—no per-CAPTCHA cost |
| CapSolver | $12 | $120 | $1,200 |
| 2Captcha | $14.50 | $145 | $1,450 |
| Anti-Captcha | $10 | $100 | $1,000 |
| CapMonster Cloud | $3 | $30 | $300 |
| AZcaptcha | $9 | $90 | $900 |
| EndCaptcha | ~$44 (package) | ~$479 (package) | ~$4,790 (linear) |
| Best Captcha Solver | $8 | $80 | $800 |
For reCAPTCHA v2 at 1M solves/month, the spread is roughly $600 (CapMonster at $0.60/1K) to $2,990 (2Captcha at peak dynamic pricing). That's a 5x difference just from choosing a different provider.
Hidden Costs to Watch
The per-solve price is never the full picture:
- Proxy costs can dwarf solver costs. Oxylabs residential pricing ranges from $6/GB to $1.50/GB depending on volume. Decodo charges $8.50/GB pay-as-you-go.
- Failed solves that get refunded still burn proxy bandwidth, browser time, and session reputation.
- Retry logic and queue management require developer time to build and maintain.
- Browser runtime costs CPU/RAM, especially at scale.
- CAPTCHA provider updates can break AI solvers overnight.
Thunderbit's model sidesteps all of these for its target users. You pay per output row of extracted data. No proxy bills, no solver queue, no retry logic.
Which CAPTCHA Solvers Handle High-Volume Parallel Scraping
Concurrency is the pain point I see most often in scraping forums. One Reddit user put it bluntly: "When we run 3 workers in parallel, we start hitting Cloudflare CAPTCHA." Another described "At ~50 nodes running in parallel, rate limiting is constant."
So which solvers actually handle concurrency?
| Service | Concurrency Signals | Practical Note |
|---|---|---|
| Thunderbit | Cloud scraping: up to 50 pages in parallel | Sidesteps token queueing entirely |
| CapSolver | Rate-limit errors documented; 120s task timeout | AI-first; use backoff on rate limits |
| 2Captcha | Capacity: 11,273/min for reCAPTCHA v2, 2,883/min for Turnstile | High capacity, but poll every 5s after 10-20s wait |
| Anti-Captcha | "Unlimited parallel processing threads"; queue stats API | Query queue load before bursts |
| CapMonster Cloud | Poll every 2-5s; max 120 getTaskResult requests per task | Fast AI-first; sane polling required |
| AZcaptcha | No hard public concurrency cap found | Test burst behavior on your target |
| EndCaptcha | Claims "tens of millions of submissions per day" | SLA strongest for covered image types |
| Best Captcha Solver | No stable public concurrency cap found | Human-worker service; test before 500-task bursts |
What happens when you send 500 requests at once? AI-based services (CapSolver, CapMonster, AZcaptcha) generally accept and return faster for supported types, but fail quickly on unsupported variants. Human-worker services (2Captcha, Anti-Captcha, BCS) may queue, return no-slot errors, or show rising latency as worker availability becomes the bottleneck.
The production answer is a rate-limited job queue with callbacks, per-provider circuit breakers, budget caps, and fallback routing. Or, if you just need the data: use an avoidance-based tool like Thunderbit and skip the queue entirely.
Self-Hosted and Open-Source CAPTCHA Solvers: An Honest Look
"Can I solve CAPTCHAs locally without paying a service?" The question comes up in forums constantly. The honest answer: it depends on the CAPTCHA type.
Where Self-Hosted Works
- Simple image and text CAPTCHAs: Viable with open-source OCR models. ddddocr (14K+ GitHub stars) is a solid offline OCR SDK. Hugging Face hosts CAPTCHA-tagged models including jtrecenti/txtcaptcha-crnn with 5M+ downloads.
- Some Cloudflare challenge flows: FlareSolverr (v3.5.0 as of May 2026) can handle some challenge-cookie flows, but it's fragile and target-dependent.
- Best for: Very high-volume simple CAPTCHA solving where you have GPU resources and technical skill.
Where Self-Hosted Falls Short
- reCAPTCHA v3 and Cloudflare Turnstile are essentially unsolvable locally. They rely on behavioral scoring, session analysis, and browser environment checks—not image recognition.
- hCaptcha Enterprise and FunCaptcha adapt dynamically, making self-hosted models a constant maintenance burden.
- A 2025 benchmark (Open CaptchaWorld) found the best MLLM agents topped 40.0% vs. human 93.3% on diverse CAPTCHA types—still far below production reliability.
| Approach | Simple Image CAPTCHA | reCAPTCHA v2 | reCAPTCHA v3 | Cloudflare Turnstile | Cost | Setup Effort |
|---|---|---|---|---|---|---|
| Self-hosted ML model | ✅ Viable | ⚠️ Partial | ❌ Not viable | ❌ Not viable | Free (+ GPU) | High |
| FlareSolverr | ❌ | ❌ | ❌ | ⚠️ Partial | Free | Medium |
| Paid API (2Captcha etc.) | ✅ | ✅ | ✅ | ✅ | $$ at scale | Low |
| AI scraper (Thunderbit) | N/A (avoided) | N/A (avoided) | N/A (avoided) | N/A (avoided) | Credit-based | Very low |
Bottom line: Self-hosted is niche-viable for simple, stable CAPTCHAs at high volume. For modern JS-based challenges, paid APIs or avoidance tools remain the practical choice.
How to Pick the Right CAPTCHA Solver for Your Use Case
Rather than a generic "it depends," I put together a scenario-based guide:
| Your Situation | Recommendation | Why |
|---|---|---|
| Non-technical user who just wants the data | Thunderbit | No proxies, no API, no solver queue—pay per output row |
| Scraping Cloudflare-heavy sites at scale | CapSolver or CapMonster Cloud | AI-first Turnstile support, fast, competitive pricing |
| Encountering unusual or custom CAPTCHAs | 2Captcha | Broadest type coverage, human adaptability |
| Accuracy and queue transparency matter most | Anti-Captcha | Human network plus real-time queue stats API |
| Cheapest path for high-volume reCAPTCHA/Turnstile | AZcaptcha or CapMonster Cloud | Lowest current per-1K rates for common types |
| Need SLA guarantees for mission-critical pipelines | EndCaptcha | Slowness and outage insurance are the differentiator |
| Drop-in replacement for existing 2Captcha integration | CapMonster Cloud | API compatibility, common SDKs |
| Complex visual CAPTCHAs where speed is less critical | Best Captcha Solver | Human-powered, supports Turnstile/FunCaptcha/GeeTest |
The avoidance-first principle still holds: prevent CAPTCHAs through proper request hygiene, proxy rotation, and fingerprint management, and you save money, time, and operational complexity. A solver should be the fallback, not the first move.
If you want to go deeper, the Thunderbit blog has guides on web scraping without coding and what web scraping actually is. And if you want to see the avoidance-first approach in action, give the Thunderbit free tier a try—you might find you never need a CAPTCHA solver at all.
Happy scraping—and may your pipelines stay challenge-free.
Try Thunderbit for CAPTCHA-Free Scraping Get Started Free
FAQs
1. What is a CAPTCHA solver and how does it work?
A CAPTCHA solver is a service or tool that resolves CAPTCHA challenges so an automated script can keep running. Some use AI to recognize images and generate tokens. Others employ human workers. And avoidance-based tools like Thunderbit prevent the challenge from appearing in the first place through managed infrastructure, fingerprint management, and proxy rotation.
2. Are CAPTCHA solvers legal to use?
CAPTCHA-solving tools are not inherently illegal, but legality depends on context. Users must comply with each website's Terms of Service, applicable laws (like the CFAA in the US), and privacy regulations (like GDPR in the EU). The hiQ v. LinkedIn case narrowed some CFAA theories for public data, but it didn't give blanket permission for all scraping. Always check robots.txt (RFC 9309) and use data responsibly.
3. What is the fastest CAPTCHA solver in 2026?
Among common CAPTCHA types, AI-first services like CapSolver and CapMonster Cloud publish the fastest claims—often 1-10 seconds depending on the challenge. AZcaptcha claims image CAPTCHA speeds of 0.3-1 second. But the fastest approach overall is CAPTCHA avoidance: tools like Thunderbit that never trigger challenges experience zero solve latency.
4. Can I solve CAPTCHAs for free?
A few services offer free tiers or trial credits. EndCaptcha provides 1,000 free credits, AZcaptcha offers $0.10 free credit after email verification, and CapMonster Cloud has a $0.10 test balance available on request. Self-hosted open-source models (like ddddocr) are free in software cost but require GPU resources and technical setup. Thunderbit offers a free tier of 6 pages/month that works without encountering CAPTCHAs.
5. What is the difference between CAPTCHA solving and CAPTCHA avoidance?
Solving is reactive: the scraper hits a challenge, sends it to a service, receives an answer or token, and injects it back into the session. Avoidance is proactive: the scraper uses proper request patterns, session persistence, proxy rotation, and browser fingerprint consistency so the site's risk engine never triggers a challenge in the first place. Avoidance is cheaper, faster, and more reliable. It's not always possible for every target site, but it should always be the first thing you try.
Learn More
