3,402 real releases, every one extracted from a live changelog and linked back to its source. No estimates. The first data-driven benchmark for how the software industry actually ships.
of 474 tracked
verifiable
last 7 days
We Went to the Source
Plenty of people have opinions about how the software industry ships. Almost nobody has gone and counted.
So we did. For each of 474 software companies, an agent fetched the live changelog and extracted only the release entries that literally appear on the page — verbatim titles, the real dates, and a link back to each one. The 194 companies that had no changelog URL on file went through a discovery step first, where the agent searched for and confirmed a real changelog before anything was captured. Freshness and release type were then computed in code, not guessed by a model.
This matters, because the easy version of this report is the dishonest one. It’s trivial to ask a language model to “estimate” how companies ship and get back a clean, confident 50/30/20 split that’s completely fabricated. We didn’t do that. Every number below traces to a real entry on a real page that you can click through and verify yourself.
The result is 3,402 individual releases from 359 verified companies — a snapshot of what the software industry was actually shipping as of June 2026. Here’s what the data says.
1. What the Industry Is Shipping
The headline: nearly half of all releases are net-new features. The industry is in build mode.
Two things stand out. First, features alone account for roughly half of everything shipped, while improvements and fixes split the remainder almost evenly. The picture is a software industry that talks far more about what’s new than about what got better or what got fixed.
Second — and this is the part most reports would quietly hide — 12% of entries couldn’t be confidently typed. The changelog simply didn’t say whether it was a feature, an improvement, or a fix. We left those unlabeled rather than inventing a category. That honest gap tells you something real: a lot of changelogs are written without a consistent structure that a human, or an AI agent, can reliably parse.
2. Every Category Ships Differently
Aggregate numbers hide the most interesting finding: categories have distinct personalities. What “shipping” means depends heavily on what kind of software you build.
A few patterns worth calling out:
AI companies ship features, not fixes. Of 445 AI releases, 62% were net-new features and only 9% were fixes. These are companies in aggressive expansion mode, and their changelogs read like roadmaps in reverse. Whether that reflects how they actually build or just what they choose to announce is the open question.
Marketing tools barely admit to bugs. Marketing and growth software had the most feature-heavy changelog of any category at 69% features — and just 7 fixes across 154 entries. Either these products are remarkably stable, or fixes don’t make it into the public update feed. Both are plausible.
Open source is the only place fixes win. Across 707 open-source releases, fixes (41%) outnumbered features (31%). This is the inverse of every other category, and it reflects a genuinely different relationship with users. For a framework or a library, stability is the product. Each patch release is the value being delivered, and the changelog says so plainly.
3. Most Verified Companies Ship Constantly
Of the 359 verified companies, 177 had shipped within the previous 7 days and 278 within the previous 30. When we captured the snapshot, roughly half the index had pushed something in the last week.
We capped capture at the 10 most-recent entries per company, so this isn’t a velocity race — a company shipping daily and a company shipping weekly can both fill their ten slots. But the freshness signal is clear: among companies that maintain a real, public changelog, regular shipping is the norm, not the exception. The recent feed ran the full range, from Linear — Team documents and GitHub — Enterprise Teams is now generally available to point releases like Astro — astro@6.4.4 and lazygit — v0.62.2 landing the same week.
4. The Uncomfortable Finding: Most Changelogs Aren’t Readable
Here’s the result we didn’t go looking for. Of the 474 companies we tried to track, only 359 had a public changelog that was live, real, and dated. That’s about three in four. The other quarter failed verification, and the breakdown is its own story.
Set aside the genuinely small companies. The notable group is the 30 whose “changelog” was actually a marketing blog, a press newsroom, or a features-overview landing page — and the 26 with real entries that carried no dates at all. A version number with no date is invisible to anyone trying to answer “what changed recently,” and increasingly that “anyone” is an AI assistant fetching the page on a user’s behalf.
This is the gap ReleasePad exists to close. AI made shipping faster than ever, but communicating what shipped is still manual, inconsistent, and — for a surprising share of companies — effectively broken. A changelog that humans can skim and agents can parse isn’t a nice-to-have anymore. It’s the difference between a release that registered and one that didn’t. We dug into exactly what AI coding tools see when they read your changelog — and the dated, structured entries are the ones that survive.
How This Was Made
Every company in the index was fetched live. Entries were extracted verbatim from the page, dates were read off the entry, and release type was inferred in code from the actual text — never asked of a model. What didn’t survive was cut: dead links, marketing pages posing as changelogs, and anything we couldn’t date. This first edition reports only the 359 companies that passed, and 100% of the 3,402 entries link back to their original source.
It’s a snapshot, not a leaderboard. We capped capture at the 10 most-recent entries per company, so this measures what’s shipping now across the industry rather than ranking anyone on volume. We’ll run it again, and the methodology stays the same: go to the source, count what’s really there, and don’t guess the rest.
Further Reading
- What AI Coding Tools Actually See When They Read Your Changelog — why dated, structured entries survive extraction and color-coded badges don’t.
- AI Agents Are Reading Your Changelog: What That Means for Product Teams — the rising audience behind the “undated is invisible” finding.
- Changelog vs Release Notes: What’s the Difference? — the vocabulary behind the Features / Improvements / Fixes split.
Built and maintained by the ReleasePad team · The Shipping Index is a snapshot as of . Every figure traces to a real entry on a live changelog.
The Shipping Index is compiled by ReleasePad — changelog software built by people who actually ship. ReleasePad auto-generates release notes from your GitHub commits, so your changelog stays current without becoming a second job.
Frequently Asked Questions
What is the Shipping Index 2026?
The Shipping Index 2026 is a data-driven benchmark of how the software industry ships, built by ReleasePad. An agent fetched the live changelog of 474 companies and extracted 3,402 real release entries from 359 verified companies — verbatim titles, real dates, and a source link for every entry. Nothing was estimated by a model.
How was the Shipping Index data collected?
For each company, an agent fetched the live changelog page and extracted only entries that literally appear on it. Companies with no changelog URL on file went through a discovery step first. Release type and freshness were then computed in code from the actual text — never guessed by a language model. 100% of the 3,402 entries link back to their original source.
What percentage of software releases are new features?
In the Shipping Index 2026, 49.9% of all releases were net-new features, 19.3% were improvements, 18.8% were fixes, and 12.0% couldn't be confidently typed because the changelog didn't say. Features alone account for roughly half of everything the industry ships.
How many companies have a readable public changelog?
About three in four. Of 474 companies tracked, only 359 had a public changelog that was live, real, and dated. The rest failed verification: 43 were stale, 26 had no per-entry dates, 30 were marketing or newsroom pages posing as changelogs, 9 were unreachable, and 7 had no changelog at all.
Why do open source projects ship more fixes than features?
Across 707 open-source releases, fixes (41%) outnumbered features (31%) — the inverse of every other category. For a framework or library, stability is the product. Each patch release is the value being delivered, so the changelog leads with fixes rather than new features.
How often do software companies publish changelog updates?
Among companies that maintain a real public changelog, frequent shipping is the norm. Of 359 verified companies, 177 had shipped within the previous 7 days and 278 within the previous 30 — roughly half the index pushed something in the last week when the snapshot was captured.
Ready to put this into practice?
Your changelog shouldn't be an afterthought.
ReleasePad makes it easy to publish great release notes — from a public changelog page to an in-app widget, GitHub integration, and analytics. Free to get started.
Get started — it's free