The debate over "self-hosted proxies or a ready-made Scraping API" is often settled with a single phrase: "for large volumes, proxies are significantly cheaper." While this statement is true, it doesn't address the right question. In 2026, it's not the volume of requests that determines success or failure, but rather the weight of a single response — with a tipping point around 400–550 KB. Below is an analysis of three solutions based on criteria, a fair arithmetic in dollars per 1000 successful responses, and a matrix of "what to choose for which scenario."
Three Different Products That Are Often Confused
First, let's clarify the entities — half of the failed budgets arise from teams comparing incomparable options.
- Raw Proxies. You pay for traffic — usually per gigabyte. The provider gives you a pool of IPs and rotation; everything else is your code: TLS fingerprint, header order, retries, cookies, rendering, HTML parsing.
- Web Unblocker. An intermediary layer: the same endpoint as the proxy, but inside — fingerprint matching (TLS handshake, headers, and JS environment yield the same browser), built-in JS rendering, automatic IP/header/timing changes when a challenge appears, sticky sessions, and transfer of anti-bot tokens between attempts. Payment is for successful responses or per gigabyte.
- Scraping API. You send a URL — you receive ready-made HTML or already parsed JSON. Maximum convenience, minimum control over the pipeline, payment for 1000 successful calls.
The key difference is not technical but economic: proxies charge for bytes, while APIs and most unblockers charge for events. Until you convert both into a single metric, the comparison is meaningless.
What Independent Measurements Show for 2026
In May 2026, aimultiple published a benchmark on over 30 million pages: 50+ products from six companies working with web data. For unlocking protected targets, the results are as follows: Bright Data — 98%, Zyte — 97%, Decodo — 96%, Oxylabs — 95%. Bright Data claims a 97.9% success rate for its Web Unlocker.
An important caveat from the researchers themselves: the success of proxy solutions depended on the timing of the test and the configuration of the request, rather than being a static figure — providers update their IP pools, and platforms enhance their security. The second direct conclusion from the benchmark is that for heavy targets like Amazon, unblockers are indeed necessary, while for simpler sites, regular proxies are sufficient.
This is why the question intensified in 2026. Detection has shifted from "IP reputation" to session behavior — Cloudflare, for example, evaluates the connection history even before the main request (we discussed this in our article about precursor sessions in bot detection). A clean IP is no longer a sufficient condition: if the stack cannot maintain a consistent fingerprint, expensive residential traffic gets burned on 403 errors.
Pricing on One Table
Public prices as of July 2026 (starting rates, without volume discounts):
| Solution | Payment Model | Starting Price |
|---|---|---|
| Zyte API | per 1000 requests | $0.13–1.27 |
| IPRoyal Web Unblocker | per 1000 requests | from $1.00 |
| Decodo | per 1000 requests / per GB | $1.25 / $10 |
| Bright Data Web Unlocker | per 1000 requests | from $1.50 |
| Oxylabs Web Unblocker | per GB | from $9.40 |
| Raw Proxies (ProxyCove) | per GB | $1.5 datacenter / $2.7 residential / $3.8 mobile |
The price range for Zyte is tenfold — this is not a typo: the price depends on how much "heavy artillery" (rendering, premium pool) the specific target required. This is a common characteristic of success-based models: the more complex the target, the more expensive the call.
Arithmetic: Dollars per 1000 Successful Responses
The only correct metric is the cost of 1000 successful responses. This should be calculated considering failures: unsuccessful attempts in the proxy model also incur traffic costs. Let's assume a success rate of 70% — meaning 1.43 attempts for one success.
Scenario 1: Only HTML, No Images or Scripts
A typical product page in "clean" HTML is about 200 KB. For 1000 successes, it takes 1.43 × 200 KB × 1000 ≈ 286 MB ≈ 0.28 GB.
- Residential at $2.7/GB → ≈ $0.76 for 1000 successful responses.
- Datacenter at $1.5/GB → ≈ $0.42.
- Unblocker at $1.00–1.50 per 1000 → $1.00–1.50.
Proxies are cheaper by 1.3–3.5 times. This is the very case from which the legend "proxies are always more profitable" arose.
Scenario 2: Full Page Render in the Browser
According to Web Almanac 2025 from HTTP Archive (published January 16, 2026), the median weight of a homepage is 2710 KB, an internal page is 1866 KB, and the median on desktop is 2412 KB. Thus, a full render through a proxy consumes about 2 MB per attempt.
- Residential: 1.43 × 2 MB × 1000 = 2.86 GB × $2.7 → ≈ $7.7 for 1000 successes.
- Unblocker for successful request: $1.00–1.50 — cheaper by 5–7 times.
- Unblocker charging per GB ($9.40): 2.86 GB → ≈ $26.9. The worst combination.
The conclusion that breaks conventional logic is this: for JS-heavy targets, raw proxies are more expensive than a ready-made API, and per-GB pricing of the unblocker penalizes for each render — one successful response consumes megabytes of fonts, images, and analytics.
Break-even Point
The formula is simple: divide the price for 1000 requests by the price per gigabyte. At $1.50/1000 and residential $2.7/GB, we get 0.556 GB for 1000 successes — that is, 556 KB per successful response, or approximately 390 KB per attempt at a 70% success rate. For a cheap unblocker at $1.00/1000, the threshold drops to approximately 370 KB per success.
The practical takeaway is: for average responses up to ~400 KB — use proxies; for significantly heavier responses — consider an unblocker. And here lies the most underrated lever: blocking images, fonts, and media in Playwright/Puppeteer reduces render weight by 5–10 times and brings the economics back in favor of proxies. Before paying for an unblocker, try simply stopping the download of unnecessary banners.
Hidden Cost: Engineering Time
Traffic is not the only cost. With raw proxies, you take on what the unblocker does internally:
- Consistent fingerprint. TLS handshake, header order, and JS environment must describe the same browser. Desynchronization is the main marker of a bot (tech breakdown in the article about bypassing JA4 fingerprint via curl-cffi).
- Failure classifier. Hard block (403), challenge page with code 200, timeout, empty render, truncated HTML — these are five different situations with different reactions. Counting "200 OK" as a success is a sure way to fill the database with junk.
- Retry policy. Changing IPs, headers, and timing, rather than just repeating the same request.
- Session storage. Sticky sessions and transferring anti-bot tokens between attempts instead of starting from scratch.
- Rendering power. Your own fleet of headless browsers, their memory, and crashes.
For a team that already maintains a scraping stack, this is not an expense — the infrastructure is already in place. For two developers with a one-week deadline, this is the same $1.50 per 1000 requests, only paid out in salaries.
Scenario Selection Matrix
- Large constant volume, light responses (HTML, JSON endpoints), your own stack. Raw proxies. Start with datacenter proxies — they offer the best price on unprotected targets; switch to residential proxies only where datacenter proxies are actually blocked.
- A few particularly protected targets, JS-heavy pages, small volume. An unblocker or Scraping API with payment for successful requests. Paying $7.7 in your traffic where a ready-made solution asks for $1.5 is stubbornness, not savings.
- A hybrid — what mature teams arrive at. Proxies for 90–95% of volume (cheap and under control), APIs for those few percent of targets that justify the premium.
- Multi-step scenarios: login, infinite scrolling, clicks, and forms. Neither will help here — a managed browser is needed; an unblocker only delivers HTML.
- Anti-pattern. Full render with images via an unblocker charging per gigabyte — the most expensive configuration possible.
Checklist Before Making a Decision
- Measure the average response weight on your targets — this is the main parameter, not the volume of requests.
- Measure the real success rate of your stack and multiply the traffic by the attempts coefficient (at 70% success — by 1.43).
- Compare only $/1000 successful responses. "$/GB versus $/1000 requests" is a comparison that always lies.
- Check if you can reduce weight: disable images, fonts, media, analytics; retrieve data from hydration payload instead of a full render.
- Consider the failure payment model: a success-based unblocker does not charge for retries, while proxy traffic always does.
Conclusion
"Proxies are cheaper at volume" is only true for light responses. The real dividing line in 2026 is the weight of a single response: up to about 400 KB, raw proxies win, beyond that — payment for successful requests. Calculate your two numbers (average weight and success rate), convert both options into dollars per 1000 successes, and the decision will cease to be a matter of faith. The most common way to save is not to change providers, but to stop downloading what you don’t need.