llama.cpp Finally Hit 1.0, and It Is a Promise More Than a Feature Drop
llama.cpp and ggml reached version 1.0.0 on November 13, adopting semantic versioning so breaking API changes arrive only as major-version bumps. It also removes 35 deprecated functions.

A Version Number That Actually Means Something
llama.cpp reached version 1.0.0 on November 13, 2026, about three and a half years after the project began. Both libraries shipped at 1.0.0 together: libllama, the inference engine, and libggml, the tensor library underneath it. The GitHub release went out at 10:20 UTC, with a Hugging Face blog post from the ggml-org team the same day.
The news is not a feature. It is a promise about how the project changes from now on. For most of its life, llama.cpp shipped continuous, build-numbered tags like b12359 and b12360, with the number derived from the git commit count rather than describing what changed. You could not tell from a version number whether an update would break your code. Now you can.
That matters because llama.cpp is the engine behind much of today's local AI software. If you run models on your own hardware, there is a good chance llama.cpp is doing the work, whether you know it or not. The project has more than 134,000 GitHub stars and sits under countless apps, tools, and wrappers.
The Semantic Versioning Rules
The rules are simple, and they apply to all ggml-org projects that use them. A MAJOR bump means a breaking change to the public C API in the include directory. A MINOR bump means a backward-compatible feature, a backend update, new model support, or an API addition. A PATCH is a bug fix with no API change.
The blog post puts it plainly: going forward, both llama.cpp and ggml will reflect API and ABI incompatibilities with major-version bumps. The API is the interface developers use to call the libraries. The ABI is the binary interface compiled apps rely on. From now on, a breaking change to either one gets a new major version, not a quiet commit in a Tuesday build.
There is one honest caveat. Semantic versioning is not a guarantee of zero pain. A minor release can still bring a backend update that changes performance behavior, making things faster or slower or using more memory, without touching any function signature. Version numbers tell you about the interface, not about the behavior. Read the release notes anyway.
What the 1.0 Release Actually Changes
The 1.0 release is a cleanup before the stability pledge. It removes the deprecated public API, including 35 deprecated functions, the llama_set_warmup function, and the defrag_thold option. It promotes the NextN speculative-decoding API into the main llama.h header. It adds FP8 KV caches, a host-memory cache for mixture-of-experts models, and new model support. It also hardens llama-server.
If your code calls any removed function, it will not build against 1.0.0. That is the point of the cleanup: the project is drawing a line under old API surface before promising not to move the line without warning.
The release also tidies up distribution. An official llama.cpp release tag is created only when its internal copy of ggml matches the released ggml, so system-wide installs through Debian and Homebrew stay compatible. That is the kind of detail nobody notices until it breaks, and it is exactly the kind of thing a 1.0 release should fix.
What to Check Before Upgrading
First, search your code for the removed functions. The big ones are the 35 deprecated functions listed in pull request 30565, plus llama_set_warmup and defrag_thold. If you depend on any of them, update your calls before you upgrade, or pin to the last 0.x release while you do.
Second, test performance, not just compilation. The FP8 KV cache changes memory behavior, and the backend updates in a minor release can shift speed in either direction. Run your own workload before you ship the upgrade to users.
Third, remember what 1.0 does not promise. The announcement does not say output will be identical between versions, that every GGUF file will load forever, or that command-line flags will stay fixed. The versioning rules apply to the public C API and to API and ABI boundaries. Everything else can still move, and it will.
Why This Matters Beyond llama.cpp
The local AI ecosystem has grown up fast, and it has been held together by projects that move at a speed that makes stability hard. A 1.0 release with real versioning rules is a sign that the ecosystem is maturing from a collection of experiments into infrastructure that other people build businesses on.
That is also why the test of this release is not the release itself. It is the next breaking change. If the team holds to its rules, the next removal from the public C API will arrive as version 2.0.0 rather than inside a routine update. If it does not, the promise was just marketing. Developers will be watching.
For now, the practical move is simple. Read the release notes, check for the removed functions, and upgrade on your own schedule. The Python 3.15 upgrade guide makes the same point for a different ecosystem: version jumps are manageable when you check first and upgrade deliberately. And if you are curious how open source projects evolve under this kind of pressure, our Rust Adobe clones story shows what happens when one developer decides to rebuild an entire suite from scratch.


