Social Media API 2026: Access, Limits and Costs per Platform

Access is the real bottleneck

Most social media automation plans run aground long before the first line of code, namely in the platform's approval process. Anyone who wants to post programmatically, collect comments or pull metrics negotiates with five different access regimes at once: partner programmes with a sales process behind them, app audits, daily allowances, per-request pricing and policies that penalise behaviour instead of volume. This side of the platforms shows up in no feature comparison.

This article deals only with the technical access side: who gets in, what it costs, where it stops. What you build on top of it, meaning publishing pipelines, monitoring or agents, belongs in AI and automation in social media management. The platforms' price lists apply globally, and we are not aware of any regional tariffs. What differs in the DACH region is the legal situation. As of August 2026.

Social media API compared: access, limits and costs

Three patterns repeat. Platforms with a price tag (X, Reddit), platforms with an allowance instead of an invoice (YouTube, Meta) and platforms with a doorman (LinkedIn, TikTok). The third group is the most awkward, because waiting time cannot be shortened with budget. How the platforms differ in substance, meaning by reach and suitability, is covered in the overview Social media platforms compared.

LinkedIn: the most expensive doorman in the field

LinkedIn draws a clean line between what any app may do and what partners may do. The Marketing Developer Platform is partner-gated. You apply, describe the use case and wait. LinkedIn publishes no price; third-party analyses put the entry level at $699 per month, based on partner statements and market data. That is an estimate with no list price behind it, and it carries no promise of approval. A market overview by the vendor Blotato puts the approval process at three to four months on average, four to eight weeks in the fastest case.

Below that sits the Community Management API. Its development tier allows 500 calls per app and 100 calls per member within 24 hours, BATCH_GET and webhooks are blocked, and the tier expires after twelve months at the latest. That is enough for a prototype. For a dashboard that queries 40 company pages including comments every day, it is not. Extrapolating dev tier figures to production load is one of the most common planning errors in this field.

The official interface does not hand out other people's profile data, it returns data on your own objects. Anyone pulling personal data out of LinkedIn does so through data providers with their own contracts, or not at all. How to build outreach without a grey area is covered in Running a LinkedIn outreach agent in line with the GDPR.

Meta: allowance instead of invoice

Instagram and Facebook cost nothing, they impose limits. Publishing runs through the Instagram Content Publishing API, which allows 100 API published posts per rolling 24-hour window. Stories can be published via the STORIES media type, direct messages are not part of this interface. Before the first post comes Page Publishing Authorization, an identity check for the linked page.

At Meta, versioning takes more effort than the daily limit. The Graph API runs in numbered versions, v25.0 was released on 18 February 2026 and returns additional error fields such as error_code and error_user_msg for failed asynchronous ads reports. Anyone coding against a fixed version has to read the changelog and actively follow field changes.

Meta also cracks down as soon as automation looks like manipulation. According to a report on Meta's own announcement, around 500,000 Facebook accounts were actioned for spam or fake engagement in the first half of 2025 and around 10 million profiles impersonating other people were removed. Automated behaviour is therefore a risk that exists independently of any quota.

TikTok: audit first, public reach second

The Content Posting API is openly documented, but access comes in two stages. As long as your client is not audited, every post is forced into SELF_ONLY mode, and at most five users may post within 24 hours. A test therefore looks functional and produces zero reach. The review process feels stricter than Meta's; that comparison cannot be evidenced, it is a judgement from practice.

The numerical limits are tight: 6 requests per minute per user token and roughly 15 to 25 videos per account per day, shared across all API clients that access the same account. TikTok's own guideline states typically around 15 posts per creator account per day. If a scheduling tool runs in parallel, your script shares that allowance with the tool, and neither of the two sees the other's consumption.

The behaviour filters bite before the numbers do. Error codes such as spam_risk_too_many_posts block earlier than any numerical limit. Since September 2025 the originality policy has demoted content carrying other platforms' watermarks as well as recycled uploads. Automated cross-posting is technically possible and, in terms of content, the worst option this API offers.

YouTube: units instead of dollars

The YouTube Data API counts in quota units. Google documents 10,000 units per day as the standard allowance of an API project. The most recent movement came on the upload side: on 4 December 2025, Google cut the quota cost of a video upload from around 1,600 to around 100 units. Six possible uploads per day became a hundred.

That shifts the arithmetic. Google's current quota documentation gives videos.insert and search.list a separate daily allowance of 100 calls each, apart from the 10,000 units. The old rule of thumb that search eats the budget now holds only in part: today the number of search requests stops you sooner than their unit price does. Anyone evaluating channel data regularly keeps searches scarce, works with stored video IDs and caches results.

X: two caps that have nothing to do with each other

X is the only major provider that bills API access like a cloud service. The price list states $0.001 per owned read, $0.005 per post read, $0.015 per created post and $0.200 per post with a link. The last figure is the most expensive item in day-to-day use: 200 link posts a month cost $40, the same volume without a link costs $3. So anyone sending traffic to their own site pays roughly thirteen times as much. On top of that comes a volume cap of 3 million post reads per billing cycle.

Legacy plans still exist, but only for existing customers: Basic at $200 and Pro at $5,000 per month, with Enterprise starting at around $42,000 monthly. For new developers the free tier has been unavailable since the switch in February 2026, according to the same market overview. Booking Enterprise without a solid business case is hard to justify at those amounts.

The second cap has nothing to do with money. X explicitly documents that rate limits and billing are separate, with 3,500 reads per 15 minutes per app on post lookup and 450 on recent search. So you collect 429 errors while your budget stays untouched. No integration runs stably here without retry and exponential backoff.

Reddit: approval before access

Reddit has moved access from self-service to prior review. Since the Responsible Builder Policy of late 2025, every developer needs an approval before the key is usable in production. Free access sits at 100 queries per minute per OAuth client, commercial use is reported at around $0.24 per 1,000 calls, and large contracts range between $50,000 and over $500,000 per year. Reddit publishes no official price list, these figures come from a market overview and should be treated accordingly.

For most DACH companies the API is the wrong tool here anyway, because the platform's value depends on genuine presence and cannot be manufactured through request volume. Why Reddit still belongs on the priority list is set out in Reddit marketing for business.

Scraping as a workaround is finished

Older guides recommend data providers that no longer exist. For years Proxycurl was the standard route to LinkedIn profile data and was shut down on 4 July 2025 after LinkedIn filed suit in January 2025. The operator ended the service himself and publicly cited the lawsuit as the reason.

From this follows a hard filter rule for research: any guide to LinkedIn data written before mid-2025 no longer works as a source. Scraping workarounds fail on contract law long before the technology becomes the issue. They breach the terms of service, and the platforms enforce that in court. This is not legal advice: as soon as personal data is involved, the GDPR assessment belongs in a specialist lawyer's review, not in a sprint ticket.

What officially cannot be automated

Direct messages are not part of the publishing APIs on any of the large platforms. Neither are some story functions and certain interaction formats. Instagram is the exception in one direction here, because stories run through the publishing API; DMs stay outside it all the same.

The predictable exception with its own rulebook is WhatsApp, where message sending is expressly provided for and priced per message. Terms, open rates and the EU hosting question are covered in WhatsApp Business API: costs, open rates and EU hosting.

In practice that means every automation ends at a point where a human takes over. Plan that handover from the start instead of discovering it as a gap in live operation.

Aggregator or direct integration

Publishing tools such as Buffer, Later, Metricool, Publer or Ayrshare hold the platform approvals themselves and pass a single interface through to you. Blotato starts at $29 per month for 20 accounts, although API access is not part of the entry plan. On the read side, providers such as Phyllo or Coresignal act as a data layer.

The trade-off is clear. You give up control over error paths, data storage and downtime, and in return you get an operation someone else looks after. No tool lifts you above the platform limits, because the allowances are tied to the account. As a rule of thumb: as long as you only publish and need no reporting of your own, a direct integration rarely pays off. As soon as your own data models, approval workflows or agents come into play, the balance tips.

Common mistakes in API planning

The mistakes in this field repeat, and almost all of them sit in the planning, not in the code. The most expensive one is approval time: at LinkedIn and TikTok the calendar decides the date, budget accelerates nothing. A project with an eight-week runtime whose plan contains a LinkedIn partner application has already failed at kickoff. Almost as common is extrapolating the development tier: 500 calls a day in development mode say nothing about production load, and the tier runs out on time anyway.

Three further patterns show up regularly:

• Rate limit confused with budget: at X both boundaries are documented separately. Anyone monitoring only the balance sees 429 errors first in the log.
• One account, several clients: TikTok splits the daily limit across all applications that access it. An extra scheduling tool may well halve your allowance.
• Behaviour filters underestimated: spam_risk blocks, originality rules and Meta's spam enforcement bite independently of any numerical limit.

What this means for planning

Treat API access like a procurement, not like a feature. Two variables decide feasibility: the approval time of the most restrictive platform in your set, and the hardest daily limit measured against your planned volume. If the first is longer than your project timeline or the second smaller than your requirement, then you have to change the architecture. The platform is no longer the thing to blame.