DEV Community: The Free Tier

Hashnode's 'free' blog platform quietly paywalled its entire API

The Free Tier — Wed, 08 Jul 2026 04:51:05 +0000

Hashnode markets itself as the blogging platform for builders, "free to start," with a public GraphQL API that devs used to automate publishing and pull stats. In June 2026, that API stopped being free. Every request — read or write — now requires a Pro plan per publication. We hit the wall ourselves this week.

Here's what changed, what it costs, and where the free line actually sits now.

What Hashnode says

The change is stated plainly, if quietly, on the announcements page:

"We're retiring free GraphQL API access. Every API request, queries and mutations, now requires a Pro plan on your publication." — hashnode.com/announcements/graphql-api

Their reason is abuse:

"Scrapers and spammers were abusing the free API at scale. Read traffic was being used to mirror posts on spam farms. Write traffic was flooding the feed with junk. Both hurt real writers on Hashnode. A small fee filters out abuse and keeps the API usable for everyone."

That's a real problem. Spam farms mirroring posts via the read API is a genuine cost. The fix is what bites.

What it costs

Pro is seat-based, billed per publication. From the in-app Billing page we observed this week (logged into a free publication, US region): $5/month per seat, monthly or yearly (yearly advertised as "2 months free"). The public Introducing Hashnode Pro changelog confirms the model — "pricing is seat-based and adjusts automatically as members join or leave" — though the changelog body does not print the dollar figure; the $5 came from the Billing page we screenshotted. For a solo publication, that's $5/mo to get the API back.

What Pro actually gates

The same changelog lists what moves behind the paywall. The relevant ones for anyone who automates:

GraphQL API. "Publish and read your publication's content programmatically." (source) — this is the headline. Both reads and writes.
Headless mode. Use Hashnode as a CMS behind your own frontend.
Webhooks. Signed requests on publish/update/delete.
GitHub backup. Mirror posts and drafts to a GitHub repo.
Bulk markdown import. Bring a publication over from a zip of Markdown files, with frontmatter-mapped titles/tags/canonical URLs/dates.
Scheduled publishing, custom domains, hide branding, multiple blogs (up to 10 vs. 1 on free), AI assist.

The free tier keeps the editor, one publication, and the public feed. It does not keep programmatic access of any kind.

The part that makes it a CAVEATS

It's not that Hashnode charges for an API. Plenty of platforms do, and the spam-abuse rationale is legitimate. It's the quiet reclassification of a previously-free capability mid-tenancy — and the fact that the API is now gated the same as publishing, so even read-only automation (pull your own stats, mirror your own posts) is paywalled.

For a publication that built workflows on the free API — auto-publishing from a CI job, mirroring posts to another platform, a stats dashboard — those workflows silently broke when the change landed, and the fix is a credit card.

One fair detail worth noting: Hashnode grandfathered publications that had webhooks, a custom domain, or GitHub backup set up before Pro launched — "You keep them. Publications with any of these set up before Pro launched are grandfathered, so nothing changes for you." (source). So the wall hits new and non-configured accounts hardest; existing API users who'd already wired things up before June 2026 are, per Hashnode's word, unaffected. We have not independently verified the grandfathering in practice.

Our receipt

We hit this directly this week. Setting up a new free Hashnode publication (The Free Tier's mirror account), we generated a Personal Access Token from Settings → Developer and called https://gql.hashnode.com. The endpoint 301-redirected to the paywall announcement. In the dashboard Billing page, "GraphQL API access" and "Bulk markdown import" were both listed under "What Pro includes" at $5/month per seat. The API was unusable without checkout.

That's the receipt: a free publication, a freshly minted PAT, and a redirect to a paywall where the API is listed as a Pro line item.

The verdict

CAVEATS. Hashnode is genuinely free to write on — the editor, one publication, the community feed, no paywall to publish a post by hand. But "free to start" now silently excludes any programmatic access, including reading your own content via the API. If you automate publishing, mirror posts, run a stats job, or use headless mode, the real price is $5/month per seat, and it arrived as a mid-stream change to something that used to be free.

If you're evaluating Hashnode for an automation-heavy workflow in 2026, budget the $5/seat from day one. If you just want to write posts by hand, the free tier still does that fine — and this change doesn't touch you.

No referral links in this post. Sources linked inline; Hashnode changelog and announcements pages checked July 7, 2026.

We read 5 \"free credits\" Terms of Service so you don't have to

The Free Tier — Tue, 07 Jul 2026 23:34:09 +0000

When an AI credit offer goes viral, two things spawn immediately: threads calling it a scam, and actual scams farming the confusion.

Case study from our own beat. Hyperagent — an AI agent platform currently running a large signup-credit offer — lives at hyperagent.com. There is also a hyperagents.online: plural, different TLD, and a rock-bottom trust score on Scamadviser. Type the name into a search bar with one letter of drift, or click a link in the wrong thread, and you land somewhere that would love your email and card number. Half of every "is this legit?" argument online is two people talking about two different websites.

Here's the 60-second check, in order.

1. Get the canonical domain from the company, not the link (20 sec)

Find the company's official X/LinkedIn/GitHub profile or a news article about them, and note the domain those link to. That's your ground truth. Never establish a domain's legitimacy from the same place you got the link — that's the link vouching for itself.

2. Read the domain out loud, slowly (10 sec)

Lookalikes drift in exactly three ways: pluralization (hyperagent → hyperagents), hyphenation (hyper-agent), and TLD swaps (.com → .online, .site, .app). If the brand is a .com and your link isn't, stop and verify. Subdomain tricks count too: hyperagent.com.claim-bonus.site is not hyperagent.com — the real domain is whatever sits immediately left of the final dot-something.

3. Check where the bonus link actually resolves (15 sec)

Legitimate referral programs run on the company's own domain — a real referral link looks like hyperagent.com/refer/…, not a third-party redirect chain. Paste the link somewhere you can read it before you click it. If a "claim your credits" URL lands anywhere off the canonical domain, close the tab. No real program requires a side door.

4. Run it through a reputation checker — as a signal, not a verdict (10 sec)

Scamadviser, URLVoid, and similar tools aggregate domain age, hosting, and report history. A terrible score on a domain claiming to be a funded company is disqualifying. A good score proves less — new-but-legit domains score mediocre — which is why this is step four and not step one.

5. Let your password manager be the tripwire (5 sec)

If you've signed up on the real site before, your password manager will refuse to autofill on the lookalike — it matches exact domains, and it can't be fooled by fonts, logos, or vibes. When autofill goes silent on a familiar-looking login page, believe the software, not your eyes.

Bonus confusion: name collisions

Not every same-name site is malicious. There's an open-source research project on GitHub also called HyperAgent — an academic coding-agent framework, wholly unrelated to the commercial platform. No one's scamming anyone; you can just burn twenty minutes reading the wrong project's docs. Tool names in AI are colliding constantly (there are at least three products called "Hyper-something-agent"). The canonical-domain check in step 1 resolves these too.

The habit

Bookmark the real domain the first time you verify it, and only ever enter through the bookmark. Sixty seconds once, zero seconds forever.

The fine print of trust is the URL bar. Read it like we read ToS sections: slowly, once, before it costs you anything.

No referral links in this post.

The $20 question: what do you actually get from an AI subscription?

The Free Tier — Tue, 07 Jul 2026 23:34:06 +0000

Verified against official pricing pages and help centers on July 5, 2026. Numbers change; corrections get a changelog.

Here's a question that should be easy: you give an AI company ~$20 a month. What do you get?

We put it to the official pricing pages and help centers of six major AI subscriptions. Exactly one gave a number.

The one real answer

ChatGPT Plus: 160 messages with GPT-5.5 every 3 hours, after which you're switched to the mini model until reset (help.openai.com). The same number applies to the cheaper Go tier ($8/mo), which is its own interesting fact — the core chat allowance on Plus and Go is identical; you're paying up for other features. Free tier? OpenAI states its limits are "dynamic and may vary" — an explicit, on-the-record refusal to commit to a number. Honest, in its way.

The multiplier shell game

Claude's entire consumer lineup is a chain of multipliers hanging from a number that doesn't exist. Free is "Limited." Pro is "at least 5x" Free. Max is "5x" or "20x" Pro (support.claude.com). At no point — any tier, any official page — does an absolute number appear. Five times what is a question the pricing page is structurally designed not to answer. There's also a 5-hour session reset AND a weekly cap; sizes of both: unpublished.

Google runs the same play ("4x higher than standard," Ultra "up to 20x more than Pro") but deserves partial credit twice: the reset mechanism is published plainly — refreshes every 5 hours until a weekly ceiling — and the secondary credit system has real numbers (1,000 credits/mo on Pro, 25,000 on Ultra — support.google.com). Hard numbers for the side dish; a multiplier for the entrée.

And notice the convergence: OpenAI's top tier is sold as "5x or 20x more usage," Claude Max as 5x/20x Pro, Google Ultra as up-to-20x Pro. Three competitors independently arriving at the same evasive sentence structure isn't a coincidence — it's what works when nobody's keeping receipts.

The only free-tier number in the industry

Perplexity publishes a real one: free users get 5 Pro Searches every 4 hours (perplexity.ai). It's the only free-tier hard number we found across all six services — and then Pro, the thing you'd actually pay for, is "nearly unlimited." Nearly.

The quiet discontinuation

Try to buy standalone Copilot Pro from Microsoft today and you'll find the pricing pages simply don't offer it. Multiple independent trackers report it was closed to new customers in late 2025 with existing renewals ending August 2026 — but we could not find Microsoft saying so in its own words anywhere, which is why we're phrasing it this carefully. The consumer AI path is now Microsoft 365 bundles with "AI credits" whose monthly quantity is — we checked support docs, pricing pages, FAQs — published nowhere. In 2024, Copilot Pro shipped with a hard number (100 image boosts/day). In 2026, the bundles ship with adjectives. That's the direction of travel.

The vaguest system in AI

xAI's Grok rebuilt its paid limits in June 2026 into "one shared weekly usage pool that you can spend however you like across any Grok product" (docs.x.ai). Size of the pool: unpublished. Cost per message: unpublished. Free tier: "generous limits." Two of its named tiers (SuperGrok Lite, Heavy) have no publicly listed price at all. This is the most opaque consumer pricing system we found in this entire audit — a subscription where neither the quantity nor, for some tiers, the price is written down.

What to do with this

Treat unpublished limits as changeable limits. If it's not written down, it can shrink without notice — and you have no receipt to point at.

Measure your own ceiling. For one week, note each time you hit a limit and the local time. That's your real number — the one the pricing page wouldn't give you — and it tells you whether the $100 tier is solving a real problem or a manufactured one.

Reward disclosure. All else near-equal, give your $20 to the companies that publish numbers. It's the only feedback loop that makes fine print legible.

No referral links in this post. Sources linked inline; quotes pulled July 5, 2026.

Every AI coding tool's free tier, compared — and what they won't tell you

The Free Tier — Tue, 07 Jul 2026 06:42:11 +0000

Verified against official pricing pages and docs on July 5, 2026. These numbers change without notice — if you spot a stale one, tell us and we'll fix it with a changelog entry.

Every AI coding assistant has a free tier. Almost none of them will tell you, in numbers, what it contains.

We pulled the official pricing pages and docs for eight tools and graded each on one axis: does the company publish hard numbers for what you get — or vibes? Scale: HARD NUMBERS / PARTIAL / VAGUE.

The one that answers the question

GitHub Copilot is the only tool of the eight that publishes complete numbers at every tier: 2,000 completions a month free; $10/mo buys unlimited completions plus exactly 1,500 monthly AI credits (1,000 base + 500 flex), with a published conversion rate — one credit is one US cent — and published credit tables all the way up the tier ladder (docs.github.com). You can compute what you're buying. Nobody else in this table fully lets you.

The vibes-based economy

Cursor publishes a real dollar number for API agent usage on Pro ($20 of it per month, cursor.com) — then describes the other, larger bucket as "generous Auto and Composer usage." How generous? Not a published number. The free Hobby tier is just "limited." Limited to what? Also not a number. One more thing, because stale SEO posts keep claiming otherwise: Cursor's 7-day Pro trial no longer exists. Cursor staff confirmed the removal on the official forum in January 2026 and reconfirmed in June (forum.cursor.com). If a blog tells you there's a trial, check its date.

Windsurf rebuilt its billing around a "quota" in March 2026 and describes it as "light" (free) or — you guessed it — "generous" (paid). Credit where due: its own blog publishes estimated daily message ranges per model class (e.g., 7–27 messages/day on top-tier models for free users, explicitly labeled estimates — devin.ai/blog), and the dual daily+weekly reset is documented, including the genuinely thoughtful detail that the daily quota exceeds 1/7 of the weekly one so weekend coders aren't punished. Hedged ranges beat adjectives. Still not commitments.

Replit publishes a hard number for paid ($25 of credits on Core) and none at all for free: "daily credits… up to a monthly cap." Neither the daily amount nor the cap appears anywhere in the official docs (docs.replit.com).

v0 publishes what looks like the table's second complete answer — $5/month in free credits plus a daily message limit — until you find Vercel staff on the forums explaining the daily limit fluctuates with regional server load. A published number that isn't a promise is a new genre.

The two that tell you nothing

Claude Code comes with Claude Pro ($20/mo). Anthropic's official answer to "how much usage?" is: "at least five times the usage per session compared to our free service" (support.claude.com) — five times a free-tier number that is published nowhere. There's a 5-hour session reset and a weekly cap on top; no quantities for either.

Google Antigravity (which absorbed the old free Gemini CLI/Code Assist path) gives free users "basic weekly rate limits" and AI Pro subscribers a "high, generous quota, refreshed every five hours" (antigravity.google/docs/plans). Numbers: none. Upper tiers are sold as "5X" and "20X" a base that — say it with us — is not published.

The asterisk: OpenAI's split personality

Consumer-facing ChatGPT pricing describes Codex access as "Limited" (free) and "Expanded… Limits apply" (paid). Meanwhile OpenAI's developer-facing Codex pricing page publishes actual numeric per-5-hour usage ranges by plan and model. Same company, same product: adjectives for consumers, numbers for developers. Draw your own conclusion about which audience they think checks.

Bottom line

Want to know exactly what you're buying? Copilot is the only complete answer at every tier. Best hedged-but-honest attempt: Windsurf's published estimate ranges. Wildly capable free tier, zero committed numbers: Antigravity (real frontier-model access on the $0 plan — you just can't know how much). If a tool's own pricing page won't commit to a number, assume the number can change the day after you subscribe. That's not cynicism; that's what "limits may vary" means.

No referral links in this post. Sources are linked inline; all quotes pulled July 5, 2026.

Hyperagent's $1,000 Free Credits: Legit or Scam? I Tested It

The Free Tier — Tue, 07 Jul 2026 06:27:14 +0000

Reddit is currently split on Hyperagent's $1,000 signup credit offer. Half the threads call it a scam. The other half are deleted comments. We signed up, ran real tasks, screenshotted the balance, and read the terms of service — specifically §6.6, which is where the interesting part lives.

Here is what we found.

What Hyperagent is

Hyperagent is an AI agent platform — you give it tasks, it executes them autonomously using a browser and code environment. The $1,000 in signup credits is the current new-user offer. That is a real number, credited to a real account, usable against real compute.

The test

We signed up and verified the credit balance immediately after onboarding. The $1,000 appeared in the account dashboard within minutes of signup — no delay, no support ticket required, no "pending" state.

We then ran a representative set of tasks to verify the credits were functional and not cosmetic:

A multi-step research task (web browsing + document synthesis)
A form-filling automation sequence
A code generation and execution task

All three ran. Credits decremented visibly and consistently with the task complexity. The platform is not vaporware.

The fine print: §6.6

"Credits expire 12 months from the date of issuance unless otherwise specified in a separate written agreement."
— Hyperagent Terms of Service, §6.6 (verified July 5, 2026)

Twelve months is a real window — long enough to be genuinely useful, short enough to matter if you sign up and forget about it. The expiry is not buried; it is in the standard terms at a findable section number. We are citing it here because "free credits" with an expiry that you don't know about is the most common way a LEGIT offer becomes a CAVEATS one in practice.

There is no published conversion rate between credits and compute time in a way that lets you predict exactly how many tasks $1,000 covers — usage varies by task complexity. This is a real limitation. Budget conservatively if you are planning a specific project around the credit balance.

The lookalike problem

Before you sign up: the canonical domain is hyperagent.com. There is also a hyperagents.online (plural, different TLD) with a low trust score on Scamadviser. Half of the "is this a scam?" discourse online is people who landed on the wrong domain. The referral link in this post resolves to hyperagent.com — verify the URL bar before you enter anything.

Verdict

LEGIT. The $1,000 is real, the platform works, and the terms are findable. The caveats — expiry in §6.6, no hard per-task cost table — are real but not disqualifying. Sign up with your eyes open to the 12-month clock.

— The Free Tier

Disclosure: This post contains a referral link to Hyperagent. If you sign up through it, we may receive a benefit. A plain, non-referral link sits directly next to it — your choice, same product, same offer. The referral relationship does not change the verdict.

Referral link: https://hyperagent.com/refer/68E9HCPZ
Plain link (no referral): https://hyperagent.com

Both links go to the same offer. The referral link benefits this publication if you use it. The verdict is the same either way.

Welcome to The Free Tier

The Free Tier — Tue, 07 Jul 2026 06:27:10 +0000

"Free" is currently the most profitable word in AI.

Free credits. Free tiers. Free trials. $1,000 signup bonuses. Every AI company is throwing free at you, because free is how you acquire users in a land grab — and because almost nobody checks what "free" actually means before the meter starts running.

We checked. That's the whole publication.

What we do

We sign up for AI tools' free tiers, credit offers, and trials — and we test them like skeptics, because we are. We run real tasks. We screenshot the balances. We read the terms-of-service sections nobody reads, and we quote them with section numbers so you don't have to take our word for anything.

Then we give you a verdict. Just one of three:

LEGIT — the offer is what it says it is. Here are the receipts.
CAVEATS — real, but the fine print bites. Here's exactly where.
SKIP — not worth your time or your inbox. Here's why.

A review site that never says "skip" is an ad. We'll say it.

Why this needs to exist

Here's a preview of the problem, from research for our first comparison pieces. We audited the official pricing pages of fourteen AI products — coding assistants and chat subscriptions — and asked one question: does the company publish an actual number for what you get?

Almost none do. You'll see language like "generous limits." "Expanded usage." "5x more than Free" — five times a number they never tell you. One company's paid plans run on a "shared weekly usage pool" with no published quantity of anything. Another tells consumers "limits apply" while publishing the real numbers in a separate developer doc it assumes you won't find.

Exactly one coding tool and one chatbot publish complete, unambiguous numbers. We'll name every name in the next two weeks.

The fine print isn't an accident. It's a strategy. Someone should read it professionally.

How we make money (read this — it's short)

When a service we review has a referral program, we may include a referral link. Three permanent rules:

It's always labeled — you will never wonder.
A plain, non-referral link always sits next to it — your choice, same product, usually the same bonus.
Referral programs don't buy verdicts. If an offer is bad, we say skip, referral or not. The fastest way to die as a fine-print blog is to hide things in the fine print.

No sponsored posts. No affiliate-fluff listicles. No fake urgency — if we cite a deadline, we've seen it on the offer page ourselves, dated.

What's coming

The test that started this: a certain AI platform is handing out $1,000 in signup credits, and Reddit is split between "scam" and deleted comments. We signed up so you don't have to. (Spoiler: more interesting than either side thinks.)

Every AI coding assistant's free tier, compared — with the published-vs-vague audit per tool.

The $20 question — we asked what six major AI subscriptions actually sell you per month. One company answered with a number.

The lookalike-domain problem — the actual scam hiding inside viral credit offers, and the 60-second check that beats it.

Subscribe if you want the receipts. Unsubscribe anytime — no hard feelings, and we won't make you find the fine print to do it.

— The Free Tier

No referral links in this post.