Bright Data alternatives: the cheapest plan buys 839,000 profile reads

Most people searching for a Bright Data alternative are not looking for a cheaper gigabyte. They are looking for a way out of a monthly minimum they cannot justify yet.

That is a different problem, and almost every comparison answers the other one. So here is the question I think matters: what does the floor actually buy?

To answer that you need to know what a page weighs. Nobody publishes that, so I measured it.

What Bright Data charges, read today

Read at source on 28 September 2026, and the first thing worth reporting is that the same product is quoted at three different rates on their own site:

And the commitment tiers, from that second page:

Every plan on that page carries a 50% OFF badge and a coupon code. The "from $2.50" on the main pricing page is the best rate at the top of the volume curve, not what a new buyer pays; pay-as-you-go is $4 discounted, $8 list. If you are budgeting from the number on the front page, you are budgeting from someone else's rate.

All of the above is what the pages showed on 28 September 2026. The date is doing real work in that sentence, and by the end of this post you will see why.

Our fuller breakdown of their products and who they suit is on the Bright Data page.

What a profile actually weighs

Ten platforms, one public profile read each, 8 samples per platform (3 on the last two), from a residential IP without a proxy, 10 September 2026:

* TikTok is not a live measurement. 0 of 8 attempts returned a usable page — HTTP 200, 1,462 bytes, and the anchor field absent. The 89 KB comes from a 7 September fixture, n=1. The cause is recorded as unknown; an earlier explanation we published was wrong and has been retracted.

Three things about how to read that table, because they change what it means:

• These are compressed bytes — what a proxy bill sees — estimated with a zstd level-1 estimator, not read off a meter. Level 1 compresses less than a real server's level 3, so the number errs high on purpose. It is a ceiling, not an invoice.
• From a residential IP, no proxy. The body is what a proxy would forward; the exit IP changes whether the page arrives, not how big it is.
• Every sample was validated against the same anchor field the parser uses. A response without the anchor was discarded rather than averaged in — otherwise a "profile not found" page enters your dataset as a very small profile.

So what does $499 buy?

$499 includes 141 GB. Divide:

The floor buys 839,000 reads of the most expensive platform there is. If your job needs five thousand profiles a month, you are buying 834,000 you will never touch — and that is at the worst platform. On GitHub the same money buys a quarter of a billion.

Bandwidth is not what makes this expensive. The floor is. No alternative "fixes" a per-GB rate you were never going to hit.

The finding I did not expect: the number has a shelf life

I re-ran the same probe on 26 September, sixteen days later, same method:

Nothing on our side changed. Threads simply got 10% heavier in sixteen days, and under per-GB billing that is a 10% price increase nobody announced.

The spread between heaviest and lightest moved with it: 287× on 10 September, 314× on 26 September. Both of those reproduce from their own measurements — divide the two numbers in each column and you get them. A third figure circulating in our own notes, 321×, does not reproduce against the table it sits next to, so it is not in this post. If our own number cannot be re-derived from our own data, it does not get published.

That is the real lesson for anyone budgeting against page weight, including us: a third party's page size is a measurement with an expiry date, and the date next to it is not decoration.

What I did not measure, and it matters

Bandwidth is one line of a scraping bill and not usually the biggest one. This post does not measure, and does not claim anything about:

• Proxy and IP rotation cost beyond the per-GB rate itself.
• Retries and failures. You pay for the bytes of the requests that come back wrong, and those do not appear anywhere above.
• Ban engineering — the ongoing work of staying readable as platforms change their defences. On Threads that means browser-equivalent TLS negotiation; a plain HTTP client gets 268 KB, status 200, and no counter object at all. The heavier request is the one that works.
• Parsing and maintenance of one extractor per platform, which is the thing most people are actually buying when they buy a structured API.

If someone tells you bandwidth is the main cost of scraping, they are selling bandwidth. It is the part I can measure precisely, which is exactly why it is the part I published.

The alternatives, and who each is actually for

The fifth row is free money for a certain kind of buyer. Bluesky, Mastodon and GitHub all expose public profiles through official APIs that need no key and no account — which is exactly why they sit at the bottom of the weight table. If those are your platforms, you do not need a vendor at all. Our Bluesky, Mastodon and GitHub routes read precisely those endpoints, and we will say out loud that you can call them yourself.

The last row is the one people skip. If the question behind the job is is this rising, you need the same number measured repeatedly over time, which no live fetch gives you retroactively. That is the archive, and it is one query rather than a million reads. The same argument applies to what X now charges per post read.

Where Bright Data beats us, and I'll say the number

At their promotional $0.75 per 1,000 records, the Web Scraper API is cheaper per record than our $1.80 per 1,000 credits. Their number against ours, read the same day.

Two things sit beside that and neither cancels it. The rate carries a 50% badge with no stated expiry, so it is not a price to plan a year around. And a record is not a credit — they return a page you still have to parse, while a credit here returns a normalized profile in the same shape across platforms, with fields that do not exist marked null instead of zero. Whether that parsing work is worth the difference is your call. On the raw number, they win.

Bright Data also won the case that shapes this market: Meta Platforms v. Bright Data (2024), where the court held that Facebook and Instagram's terms do not bind someone reading public data while logged off, because a logged-off visitor is not a "user." Everyone in this category benefits from that, including us.

How to decide, in one question

Is the floor the problem, or the rate?

If the rate is the problem, you are at volume and the per-GB vendors are already your answer. Take the weight table, divide by your own volume, and negotiate.

If the floor is the problem — and for most people typing this phrase it is — per-GB pricing was never the thing you needed. You needed a vendor with no minimum, a platform with an open API, or a stored history instead of a live fetch.

Our prices are on the pricing page: one credit per call, credits do not expire, a call that finds nothing is not charged, and the free key is 1,000 credits a month with no card. We sell in this category, so treat that paragraph as a vendor talking and check the arithmetic above instead — that part does not require trusting me.

Every vendor we have measured, with prices and dates, is on the comparison index.

Bright Data prices read from brightdata.com/pricing and brightdata.com/pricing/proxy-network/residential-proxies on 28 September 2026 — primary source; the overlapping figures were cross-checked against a reading of the same site on 26 September and were identical. Every plan displayed a 50% discount badge on the day of reading. Page weights are our own measurement: 8 samples per platform (3 on the last two), residential IP without proxy, compressed bytes estimated with a zstd level-1 estimator — a ceiling, not a meter reading — validated against the parser's anchor field, 10 September 2026, re-run 26 September 2026. TikTok is a 7 September fixture, n=1, not a live measurement. Per-GB arithmetic uses 1 GB = 1,000,000,000 bytes, the vendor convention. Verify before you commit engineering time.