Linus Torvalds released Linux 7.4-rc1 on November 1, 2026, closing the largest merge window in the history of the Linux kernel. The window took in almost 20,000 commits, compared with the more usual figure of about 15,000, and Torvalds said the surge is, in large part, a sign that AI triggered work is now arriving during normal development instead of piling up as fixes during the stabilization period (Linus Torvalds, Linux Kernel Mailing List, retrieved 2026-11-02). That is the headline, but it is not the whole story. The release candidate also brings new hardware support from AMD, Apple, Google, Intel, and Qualcomm, and it is widely expected to become this year's long-term support kernel, although that designation has not yet been formally confirmed (Phoronix, retrieved 2026-11-02).
Key Takeaways
Linux 7.4-rc1 was released on November 1, 2026, closing a merge window of almost 20,000 commits, the largest on record for the Linux kernel.
Torvalds links the size of the window to AI triggered work and to expectations that 7.4 will be the next long-term support release, and he calls the new pattern a much healthier model, with a clear warning that the release candidates will show whether it holds.
About a third of the patch, by Torvalds's own account, is AMD GPU register offsets and bit masks, so the headline commit count overstates the amount of new behaviour.
Major inclusions include AMD Ryzen Zen 6 preparation, AMDGPU HDMI 2.1 enabled by default, basic Apple M4 enablement, Google Tensor G5 and Pixel 10 support, and more Qualcomm Snapdragon X2 work.
This is a first release candidate for testing, not a stable release. If the usual weekly cadence holds, Phoronix expects a stable release around December 20, 2026.
A release candidate is easy to misread. It is not a finished product, and it is not a promise that every feature in it will survive unchanged. It is a checkpoint: new features are frozen, the work turns to testing and fixing, and a new candidate appears roughly each week until the result is stable enough to ship. Readers who run a distribution kernel will not get 7.4 from this announcement. They will get it later, through their distribution, after the stable release. What this announcement does give them is a reliable preview of what is coming.
What Is Linux 7.4-rc1?
Linux kernel is the core software that sits between a computer's hardware and everything else that runs on it. It manages the processor, memory, storage, and devices, and it provides the services that applications use. Release candidate is a test version that follows the merge window and precedes a stable release. Merge window is the roughly two week period at the start of a kernel cycle when maintainers submit new features and other changes for inclusion. Once the window closes, only fixes are normally accepted. Long-term support (LTS) release is a kernel version that receives fixes for a longer period than a regular stable release, which is why distributions and device makers prefer it as a base.
Linux 7.4 follows Linux 7.3, which was released on October 18, 2026. The 7.4 merge window therefore ran for about two weeks after that release, and Torvalds closed it on schedule on November 1, despite having been on holiday during the cycle (Phoronix, retrieved 2026-11-02). He also noted that two pull requests were still sitting in his inbox when he tagged rc1. He chose not to delay the release for them, and said they may or may not appear in rc2. That detail matters because it shows what rc1 is: a snapshot of a process that is still moving, not a final verdict on the cycle.
Aspect | Linux 7.4-rc1 |
|---|---|
Released | November 1, 2026 |
Stage | First release candidate, for testing |
Merge window commits | Almost 20,000, compared with about 15,000 usually |
Counted commits since Linux 7.3 | 19,671, compared with 16,267 for the 7.3 window |
Source tree size | 41,605,120 lines before rc1, up from 41,019,555 lines at Linux 7.3 |
Expected stable release | Around December 20, 2026, if the weekly cadence holds |
Long-term support status | Widely expected, not formally confirmed |
License | GPLv2, the licence under which the kernel is distributed |
Why the Merge Window Is So Big
The first explanation Torvalds gives is AI. He writes that he is much happier with the flood of AI triggered work going into the merge window than with the same work showing up mainly as extra fixes during the release candidate period. The logic is straightforward. Fixes that arrive early can be reviewed, tested, and revised as part of normal development. Fixes that arrive late, during stabilization, compete with the work of making the release stable, and each late change carries a higher risk of breaking something else. If AI tooling is finding more bugs and suggesting more patches, the kernel would rather absorb that volume when the tree is open to change than when it is supposed to be settling down.
The second explanation is the LTS expectation. Torvalds says plainly that a small part of the size may also come from people expecting 7.4 to be the next LTS release, which is another independent reason why some releases tend to be larger than others. Contributors and companies have a stronger incentive to land work in a release they expect to live for years. Neither explanation alone explains the whole window, and Torvalds says so himself: it may not just be the flood of AI patches. A careful reading keeps both causes in view, instead of turning a complicated cycle into a single slogan about AI.
The numbers behind the headline deserve the same care. Phoronix counted 19,671 commits since Linux 7.3, compared with 16,267 commits for the Linux 7.3 merge window and 14,395 commits for the Linux 7.2 merge window (Phoronix, retrieved 2026-11-02). That is a rise of about 21 percent over the previous window on that count, not a doubling, and it is a count of commits, not a measure of quality or of how much behaviour changed. Phoronix also measured the source tree at 41,605,120 lines before rc1, against 41,019,555 lines at the Linux 7.3 tag. The difference in detected code alone is 455,632 lines, which is why the tree is now inching toward 42 million lines in total (Phoronix, retrieved 2026-11-02). Lines of code are a crude measure, and a large part of any kernel diff is generated headers and driver data, but the direction is clear: this is a big cycle by every counting method available.
The largest single distortion in the count is one Torvalds names directly. A full third of the patch is just AMD GPU register offsets and bit masks. Those are the tables that describe hardware registers to the driver. They are necessary, they are large, and they tell a reader almost nothing about new features. If that third is set aside, Torvalds says the rest looks fairly normal in shape: about half drivers, with the rest spread across architecture updates, filesystems, tooling, and core kernel areas such as networking, Rust, BPF, and memory management (Linus Torvalds, retrieved 2026-11-02). The shortlog alone runs to 24,000 lines and 1.2 megabytes, which is why Torvalds warns readers not to try to read it in full. The summary below picks out the changes that matter to users, rather than repeating that list.
What Is Actually in Linux 7.4
On hardware, the headline changes are easy to group. AMD gets a new PerfOpt feature for Radeon integrated graphics aimed at AI and large language model workloads, Ryzen Zen 6 support squared away ahead of those next generation client processors launching, and AMDGPU HDMI 2.1 enabled by default for Radeon users connecting to modern televisions and monitors (Phoronix, retrieved 2026-11-02). Apple gets basic M4 enablement, with the important qualifier that it is not yet ready for end users. Google gets Tensor G5 system on a chip and Pixel 10 device support in the mainline kernel. Qualcomm gets further Snapdragon X2 laptop enablement, and Intel gets IPU8 support for upcoming Nova Lake laptops. Together, they show a kernel cycle preparing the ground for hardware that is arriving, rather than reacting to it after launch.
Beyond named processors, several changes affect everyday performance and reliability. EXT4 gains optional buffered iomap support. The kernel build process gets faster, which matters to everyone who compiles kernels. Memory management, scheduling, and workqueue changes landed across the cycle, and the merge summary lists updates for filesystems, networking, Rust, BPF, and virtualization, including KVM updates and a Landlock update for the kernel's sandboxing framework (Linus Torvalds, retrieved 2026-11-02). One change with a long tail is the removal of Google's C written Binder driver, with the Rust based Binder driver ready to take its place for future Android devices.
Two smaller additions show the same practical pattern. Raspberry Pi users get support for a 10 inch monitor, and virtual machine users get security tweaks that The Register highlights in its coverage of the release (The Register, retrieved 2026-11-02).
The AI Question, Stated Carefully
The most quoted part of this release is also the easiest to overstate. Torvalds does not say that AI wrote the kernel, and he does not say that AI patches are now trusted without review. He describes a release that is large and full of fixes for issues found by AI, with more and more of the patches suggested by AI too, although they often get rewritten. That sentence contains three different claims, and they should not be collapsed into one. Finding an issue is not the same as writing a fix. Suggesting a patch is not the same as landing it unchanged. Rewriting is review in action, and it is the part of the process that keeps the kernel's quality bar intact.
An independent count adds useful context, with a caveat attached. One editorial count of commits carrying an Assisted-by tag found 3,008 such commits in the 7.4 merge window, about 2.4 times the number counted for 7.3, but the same report notes that the tag was abolished partway through the 7.4 window, so the tag count is not a clean measure of AI involvement across the whole cycle. Treat that figure as one publication's tally of a disclosure tag, not as an official kernel statistic and not as proof of how much code AI produced. The official figure remains Torvalds's own: almost 20,000 commits in total, against about 15,000 usually.
There is also a counterweight inside the same cycle. A feature merged to help reduce AI generated noise from invalid testing configurations was later reverted, after a problem in the VFIO subsystem. AI tooling creates new kinds of noise as well as new kinds of signal, and the kernel's response to both is the same: test, review, and revert when a change does not hold up. The healthier model Torvalds hopes for is not a model without reverts. It is a model in which reverts and fixes happen early enough to be absorbed.
Who Should Test 7.4-rc1, and Who Should Wait
This release candidate fits best for:
Kernel developers and testers who can build rc1, run it on test hardware, and report regressions with enough detail for a maintainer to reproduce them.
Distribution maintainers and hardware vendors preparing for the stable release and for a possible long-term support base.
Owners of the newly supported hardware, such as upcoming AMD, Apple, Google, Intel, and Qualcomm platforms, who want to check whether their device is recognised and stable.
Anyone curious about the AI debate who wants to read the actual merge summary and the per subsystem changes, rather than relying on a headline about commit counts.
Wait if you need a stable system, if you run a server or a laptop you depend on daily, or if you cannot afford to debug a regression yourself. Wait also if you are choosing a long-term support base today: 7.4 is widely expected to become the LTS release for this year, but expectation is not designation, and the formal decision has not been announced. The safe path for most users is unchanged. Keep running your distribution kernel, watch the release candidates, and move to 7.4 when your distribution ships it as stable.
Frequently Asked Questions
Is Linux 7.4 stable yet? No. Linux 7.4-rc1 is the first release candidate. It opens a stabilization period of roughly seven or eight weekly candidates, and the stable release is expected around December 20, 2026 if that cadence holds. A release candidate is for testing, not for production use.
Did AI write Linux 7.4? No. Torvalds describes a release full of fixes for issues found by AI, with more patches suggested by AI, although they often get rewritten. Human maintainers still review, revise, and merge the work, and the merge window itself is the review process. AI changed the volume and the timing of contributions in this cycle. It did not replace the maintainers.
Why are there almost 20,000 commits? Two reasons are given by Torvalds himself: a flood of AI triggered work arriving during the merge window, and expectations that 7.4 will be the next LTS release. About a third of the patch is AMD GPU register data, so the commit count should not be read as 20,000 separate new features.
Is Linux 7.4 the next LTS kernel? It is widely expected to be this year's LTS kernel, because it is the last major release of the calendar year and the timing fits. That expectation has not been formally confirmed. Treat it as likely, not as announced, until the LTS maintainer says so.
What should I do if I want to try it? Test it on hardware you can spare, report regressions clearly, and keep your working system on a stable kernel. Most users should wait for their distribution to ship Linux 7.4 as a stable update. The kernel source and the release announcement are linked in The Bottom Line below.
How This Article Was Checked
The methodology behind this article is simple and traceable. We checked all 96 existing blogPost entries in Contentful on 2026-11-02, and none covered Linux 7.4 or the 7.4 merge window. The primary source is Torvalds's own 7.4-rc1 announcement on the Linux Kernel Mailing List, retrieved on 2026-11-02. Phoronix's release candidate report and its merge window measurements were used as an independent check, and The Register's coverage was used as a second check on the AI framing. Where a figure comes from one publication's count rather than from Torvalds, the article says so. Where LTS status is only expected, the article says expected, not confirmed.
The Bottom Line
Linux 7.4-rc1 is a big release, and the size is the story, but only if it is read correctly. The kernel took in almost 20,000 commits, a third of them AMD register data, and Torvalds reads that volume as a hopeful sign: AI triggered work is arriving when the tree is open, not when it is supposed to be settling. That hope will be tested by the release candidates. If the rc period stays calm, the healthier model claim gains weight. If it fills with late fixes, the claim weakens, and Torvalds himself says we will see how the release candidates look. Either way, the hardware work in 7.4 is real, the stable release is likely weeks away, and the LTS question remains open until it is announced. Start with the 7.4-rc1 announcement and the kernel source at kernel.org, or read the Indonesian version of this article. For how the kernel underneath it all works in practice, see our Docker explainer, since containers depend on the kernel features this release keeps developing.