mrkeyoor.com_
Tue 08 Sept 21:07 UTC
Open Source6 min read

LibreOffice 26.8's 1M Downloads Don't Prove an Anti-AI Surge

LibreOffice set a first-week download record while promoting its lack of built-in AI. The numbers show strong demand, but they do not explain it.

A 611-point Hacker News story turned LibreOffice 26.8 into a referendum on AI inside a few hours. By the time MrKeyoor's feed captured it on September 8, the discussion had also drawn 208 comments. That response is a strong measure of developer interest. The download data underneath it supports a narrower conclusion: LibreOffice had an unusually successful release week, while the reason for that success remains unmeasured.

The Document Foundation counted 1,031,162 downloads from its official page during the first week after LibreOffice 26.8 arrived. It called that the highest first-week total for any LibreOffice release and noted that the count excludes people who obtained the suite through Linux distribution repositories. The post does not provide the previous record, referral data, a user survey, or any other evidence connecting those downloads to the absence of generative AI.

The record is real; its cause is open

The timeline has become compressed in retellings. LibreOffice 26.8 launched on August 26 with "no artificial intelligence" in the announcement headline. The project published its record download total on September 2, then expanded its AI policy on September 3. On September 8, Manual do Usuário connected the two events in the article that reached Hacker News. The facts occurred in that order, but sequence alone cannot identify why a million people downloaded the installer.

Rodrigo Ghedin, the author of that article, presents the causal claim as his own interpretation. He writes that he is betting on the missing AI feature as the explanation, rather than reporting a finding from The Document Foundation. That distinction gets lost in a headline built around "after." The foundation's statistics post celebrates reach and thanks contributors; it does not credit the AI message for the record.

The launch clearly traveled. Alongside the download figure, The Document Foundation reported 50,061 combined views, shares, likes, and comments across Mastodon, Bluesky, X, and Facebook. Its new-features video had 20,918 views, with subtitles translated into 32 languages. Those first-week figures show that the release found a wide audience, though the combined social total mixes several kinds of engagement and says nothing about download intent.

The argument continued because the project's public download dashboard exposes trends without asking users why they arrived. In the Hacker News thread, some commenters read the monthly chart as broadly flat and treated the weekly peak as ordinary variation. Others argued that rising lows in the chart matter more than individual peaks. A third group suggested AI agents themselves may be driving LibreOffice use as a document conversion tool. These are community hypotheses, not independent measurements, and the thread contains no acquisition study that settles them.

The million-download mark still tells maintainers something useful. An official installer for 26.8 moved faster in its first seven days than the same channel had moved any earlier LibreOffice release, according to The Document Foundation's count. It does not measure downloads from Linux repositories, active installations, retention, or satisfaction. Calling it a release record is accurate. Calling it proof of an anti-AI migration goes past the available data.

What LibreOffice means by "no AI"

LibreOffice 26.8 ships without generative AI features. The foundation says documents are not sent to remote services for processing, and no part of the suite needs network access to function. It frames that design around files that may contain medical records, legal material, tenders, or student data. The release announcement makes a concrete product promise about local operation, rather than a general judgment about machine learning.

The follow-up policy sets six conditions for any default AI integration. Users must control where inference runs, and the application cannot transmit document content without authorization. LibreOffice's existing rule against usage telemetry must continue to apply. An integration must also avoid dependence on one provider, preserve generated content as structured ODF, and remain removable and absent from the interface for administrators who prohibit it. The foundation says no current integration meets all six requirements across a supported release lifecycle.

People who want AI inside LibreOffice can already install community extensions. According to the foundation's policy post, many connect to Ollama, LM Studio, or another OpenAI-compatible endpoint, allowing local inference when configured that way. The foundation neither develops nor maintains those extensions. It advises users and administrators to inspect whether an extension points to a cloud service and how that provider handles transmitted data.

That is a more precise position than the anti-AI label circulating around the release. The foundation's threshold concerns defaults, data movement, provider choice, file formats, and administrative control. Its September 3 statement also says it will assess a future approach that satisfies those rules, citing more practical local inference and improving open models. LibreOffice has left the extension route open in the meantime.

The foundation also connects product design to funding. Subscription software vendors can use an assistant to support a higher price and keep documents inside their services, it argues. LibreOffice has no subscription tiers or user data to monetize, so its coordinating nonprofit says it can wait for a suitable implementation. That is The Document Foundation's account of its incentives, not evidence about why customers choose competing office suites.

Version 26.8 has plenty besides the slogan

Development of LibreOffice 26.8 began in December 2025 and involved 206 contributors, 155 of them volunteers. Its largest block of work deals with bidirectional and complex text: Writer can detect paragraph direction when text is opened or pasted, Calc can set a cell's direction from right-to-left input, and formatting marks can expose bidirectional control characters. Those release details offer another plausible reason for a fast uptake among people whose scripts have historically received weaker office-suite support.

Writer's Paragraph Composer distributes spacing across consecutive lines instead of justifying each line in isolation. Calc gains calculated fields in pivot tables. Impress and Draw can put multiple page sizes in one document, while Impress adds named presentation sections. LibreOffice 26.8 can preserve several newer OOXML chart types through an import and export round trip even when it cannot display or edit them. The project's release notes in the announcement describe those limits, including the loss of geographic content in region-map charts.

There are migration details too. Gentium Basic and Gentium Book Basic are no longer bundled, so documents that request them will substitute another typeface unless the fonts are installed separately. Safe Mode is now called Troubleshoot Mode, and obsolete Java applet configuration has been removed. The 26.8 announcement also says the suite is available for Windows, macOS, and Linux in 120 languages.

Version 26.8 is the first LibreOffice release to show a periodic donation banner in the Start Centre. The foundation says the prompt appears around monthly or after a successful update, links to its donation page, and appears nowhere else. The same funding section says LibreOffice has no advertising, account requirement, subscription, or feature held behind payment. That change matters because a million official downloads also create a rare chance for a donation-funded project to ask new and returning users for support.

The sharper lesson for software teams

LibreOffice's six conditions read like an architecture review for optional AI. They ask where inference occurs, when data crosses a boundary, whether an administrator can remove the feature, and whether changing providers breaks the workflow. The foundation's policy turns those questions into release requirements. A team can adopt the same checks without adopting LibreOffice's decision to withhold a built-in assistant.

The extension model shifts part of that review to users. A local endpoint can keep a document on the machine, while a remotely configured endpoint can send the same material to a provider. The Document Foundation explicitly tells administrators to evaluate these plugins as third-party components because their maturity and data behavior vary. That warning in the policy post is easy to miss when the story is reduced to software that simply has no AI.

The next useful evidence will come from behavior after launch week. The download dashboard can show whether elevated demand persists, while future release posts can disclose whether the project changes its six conditions or adopts an integration that meets them. For now, LibreOffice 26.8 owns a verified first-week record and a fiercely discussed privacy position. The available numbers do not tell us that one caused the other.

We reviewed this

  1. composer — our honest review
  2. plugins — our honest review
  3. usage — our honest review

Sources

  1. LibreOffice breaks download records after declaring it has no AI features
  2. One week of LibreOffice 26.8: The stats
  3. Yes, no AI is now a feature
  4. LibreOffice 26.8 release announcement
  5. LibreOffice Download Metrics
  6. Hacker News discussion of LibreOffice's download record