TL;DR: As of September 2026, X has no free or flat-rate API tier left at any level: legacy Pro subscribers are being force-migrated to pay-per-use after September 1, 2026, the same fate legacy Basic hit back in June. Posting through the official API costs $0.015 per tweet, or $0.20 if it contains a link. The alternative is to skip the API and drive your own already-logged-in browser session instead, which is free and avoids the developer-app review, but it doesn’t make your account immune to X’s automation rules.
Say you run three X accounts for a small brand and want an AI to draft and publish a daily thread on each. The writing part is solved. The posting part is where the math gets ugly: at $0.015 a tweet with no link, three accounts posting five times a day comes to about $6.75 a month, which sounds cheap until you add reads, retries, and the fact that any post with a link costs over ten times more. Scale that to a real content calendar and the API bill stops being a rounding error.
Why posting to X got more expensive in 2026
For years, a small free tier let hobby bots post a handful of tweets a day at no cost. That ended on February 6, 2026, when X made pay-per-use the default for new developers: $0.015 per post created without a link, $0.20 per post with one, and $0.005 per post read, capped at 2 million reads a month. There is no free option to fall back on anymore.
The old flat tiers didn’t survive long past that either. Legacy Basic ($200/month) subscribers got force-migrated to pay-per-use on June 1, 2026, and on August 14, 2026, X announced Pro ($5,000/month) is next: every remaining Pro subscriber migrates automatically after September 1, 2026, at the end of their own billing cycle, per X’s migration notice on its developer forum. The mechanics matter if you’re on the receiving end: your card gets an initial $5,000 credit, it auto-recharges at $5,000 whenever the balance drops below $250, and there’s a $10,000 spend cap per cycle you can adjust in the developer portal. Full pricing numbers are in X’s own API pricing docs.
For a single creator posting a few times a day, that’s a few dollars a month. For an agency running dozens of accounts, or an AI agent that posts opportunistically rather than on a fixed schedule, the per-post fee turns automation into a metered utility you have to budget for.
Paying the API toll doesn’t make you ban-proof
Here’s the part most “how to automate X” guides skip: going through the official, paid API does not exempt you from X’s anti-automation enforcement. In March 2026, X ran a large suspension wave targeting what it calls “inauthentic behavior,” and a number of legitimate, non-bot accounts got caught in it. The pattern X is checking for is whether there’s any real manual activity on the account at all, not just whether you’re using an API key.
The rules, laid out in X’s authenticity and platform manipulation policy, are reasonably clear about where the line sits: scheduling and publishing your own original content is fine, but automating likes, follows, retweets, replies, or DMs is not, no matter which method you use to do it. Mass actions in a short window read as a bot regardless of whether the request came from the REST API or a browser click.
So the API solves a billing and rate-limit problem. It does not solve the “does this account look like a human” problem, which is the one that actually gets accounts suspended.
The other route: drive your own browser instead of the API
There’s a second way to let an AI post to X, and it sidesteps the per-post fee entirely: instead of going through X’s developer API, you drive your own already-logged-in browser session, the same one you’d use to scroll X on your laptop. There’s no API key, no developer-app review, and no per-tweet charge, because nothing is hitting X’s metered endpoints. As far as X’s servers are concerned, it’s your browser, with your cookies, doing what your browser normally does.
This is the model PublishPort is built around: an AI agent gets a local_bash tool and a one-line capability list (X/Twitter publish, search, reply, DM, and more), and it runs the actual posting through PublishPort’s ppcli CLI against your own logged-in session, on your own machine. We wrote about why running publishing locally instead of on a rented cloud server keeps the setup simpler: no server to provision, no session cookies to babysit remotely, and the account behaves exactly like it does when you’re the one clicking the button, because you are.
How this actually works, step by step
- Install the PublishPort desktop client and log into X normally, in the app’s own browser view, exactly like signing into any site.
- The desktop app exposes that logged-in session to a local relay, without ever sending your password or session token anywhere.
- An AI agent, wherever it runs, gets two tools:
list_capabilities()to see what’s available (X publish, reply, search, and so on) andlocal_bash(cmd)to run it. - The AI runs
twitter -hto see the actual command syntax, then calls something liketwitter post "..."throughlocal_bash, which executes on your machine using your real session. - The post goes out from your browser session, the same as if you’d typed it and hit the button yourself.
We cover the full agent-side setup, including the MCP connection details, in how to let an AI agent publish to social platforms.
What’s still off-limits, even from your own browser
Being honest about the edges matters more than the pitch, because overselling this is how accounts get suspended anyway.
- Volume and repetition are still signals. Posting the same content across accounts, or posting at a pace no human keeps up, looks mechanical no matter what’s driving the browser.
- Engagement automation is still against the rules. The golden rule from X’s own policy stands regardless of method: automate content and scheduling, never automate likes, follows, retweets, replies, or DMs.
- A real session removes the billing and review friction, not the need to follow the rules. Skipping the metered API gets you out of paying per tweet and out of a developer-app review queue. It doesn’t exempt the account from X’s own behavior-pattern review. Nothing here promises a suspension-proof account, and any tool that claims otherwise is overselling it.
- Platform rules can and do change. The enforcement wave that started in March 2026 ran into April before it settled, a reminder that “compliant today” isn’t a permanent state.
- Mirroring the same thread across accounts you own is its own violation, separate from automation. X’s authenticity policy lists posting substantially identical content across multiple accounts you control as disallowed multi-account behavior, whether a human or an AI hits publish. If those three brand accounts from the intro are meant to run distinct content per audience, that’s fine; copy-pasting one thread to all three isn’t.
FAQ
Does X (Twitter) still have a free API tier in 2026?
No. Since February 6, 2026, pay-per-use is the default for new developers, with no free tier to fall back on. Posting costs $0.015 per tweet without a link, or $0.20 with one. The old legacy flat-rate plans are being retired too: Basic was force-migrated in June 2026, and Pro is being phased out on the same automatic schedule after September 1, 2026.
What happens to legacy Pro subscribers after September 1, 2026?
Their subscription ends at their next billing cutover date and they’re moved onto pay-per-use automatically, no action required. X issues an initial $5,000 credit to the card on file, then auto-recharges $5,000 whenever the balance falls below $250, capped at $10,000 in spend per cycle unless you change those defaults in the developer portal.
How much does it cost to auto-post to X with the official API?
At $0.015 per link-free post and $0.20 per post with a link, a low-volume account posting a few times a day typically lands in the single digits to tens of dollars a month. The cost scales fast for high-volume or multi-account automation, which is the main reason teams look for an alternative to the metered API.
Will I get banned for using AI to post on X?
Using AI to draft or schedule original content isn’t, by itself, against X’s rules. What gets accounts suspended is automating engagement (likes, follows, retweets, replies, DMs) or posting at a volume and pace that no human matches. X’s March 2026 enforcement wave specifically targeted accounts with no detectable manual activity at all.
What was X’s inauthentic behavior ban wave in March 2026?
It was a large suspension wave aimed at accounts X’s systems flagged as having no real human activity behind them. It caught some legitimate accounts along with bots, which is a useful reminder that “I’m not technically violating a rule” and “I won’t get flagged” are not the same guarantee.
Can I automate posting to X without going through the developer API?
Yes. Driving your own logged-in browser session, the same one you already use to browse X, accomplishes the same posting outcome without an API key, developer-app review, or per-post fee. Tools built on browser automation, like the ppcli-based publishing PublishPort uses, work this way.
What automation does X explicitly allow?
X’s own authenticity policy allows scheduling and publishing your own original content through an authorized method. It draws the line at automating engagement signals: likes, follows, retweets, replies, and direct messages are off-limits to automate regardless of the tool.