Contents

llama.cpp and its tensor library ggml released version 1.0.0 on November 13, 2026, establishing a clear rule that incompatible changes to the public API or ABI require a major version bump. This release also removes deprecated functions, so developers who embed either library need to review existing calls before upgrading.

Key Points#

  • libllama and libggml now both use major, minor, and patch version numbers to indicate how updates affect compatibility.
  • Version 1.0.0 removes 35 deprecated functions from llama.h, along with llama_set_warmup and others.
  • llama-server adds the /v1/audio/speech endpoint for speech synthesis, while the Web UI gains named chat groups and a Markdown editor with input previews.
  • The versioning policy helps assess upgrade risks, but applications still need testing with the models and hardware they actually use.

What the Version Numbers Mean#

The most useful change in 1.0.0 is the compatibility policy for applications that depend on libllama or libggml. Version numbers now let maintainers determine whether an update changes the public integration interfaces.

The announced rules use the major.minor.patch format.

| Version Component | Changes Indicated | |---|---| | Major | Changes that break public API or ABI compatibility | | Minor | Backward-compatible feature and API additions, backend updates, and model support | | Patch | Bug fixes that do not change the API |

This provides a clear starting point when considering dependency updates. Major updates require a review of the work needed for migration. Minor and patch updates have a more limited stated scope, but their impact on individual applications still needs verification.

Why Both API and ABI Matter#

The API defines how application code calls a library. When a function is removed, code changes may be needed before it will compile again. The ABI defines compatibility at the compiled binary level. Breaking changes may require rebuilding software that links to the library.

This distinction matters for teams distributing applications that include these libraries. Confirming that source code compiles successfully is separate from confirming that a packaged application loads and runs with the intended library build.

The announced policy covers the public APIs and ABIs of libllama and libggml. It should not be taken as a blanket guarantee that llama-server, its Web UI, or every aspect of model execution will remain unchanged within the same major version.

What to Check Before Upgrading#

Version 1.0.0 includes API cleanup. It removes 35 deprecated functions from the public header llama.h, and the announcement also lists llama_set_warmup for removal. Applications that still call removed functions will need migration work.

Before adopting this release, check these four points.

1. Search application and binding code for calls to deprecated llama.h functions, including llama_set_warmup. 2. Check those calls against the 1.0.0 headers and release documentation to identify supported alternatives. 3. Rebuild the application and bindings against the versions of libllama and libggml you plan to use. 4. Run representative inference workloads on the models and hardware used in production.

The announcement provided does not include a complete mapping of removed functions to replacements. Check the release documentation before changing calls, and do not assume that similarly named functions behave identically.

Server and Web UI Additions#

For llama-server users, the main addition is the /v1/audio/speech endpoint for speech synthesis. It offers a concrete feature to evaluate if your application needs audio output. However, the endpoint's existence alone does not establish which models, request options, or deployment configurations will meet your requirements.

The Web UI adds two productivity features: named groups for organizing chats and a Markdown editor that previews chat input. These changes directly benefit people using the bundled interface.

Evaluate library migration and server features separately. Successfully rebuilding an application that embeds the libraries does not validate speech synthesis integration or confirm that the updated interface suits existing workflows.

What Still Needs Validation#

The development team says it is expanding automated testing across major hardware platforms with support from Hugging Face services and hardware vendors. This investment helps maintain a project that supports a broad range of hardware, but it cannot guarantee results for every deployment configuration.

The announcement does not indicate general improvements in speed or memory usage in 1.0.0. Treat clearer compatibility implications as the release's main operational benefit, and measure performance separately if it affects your upgrade decision.

Next Steps#

Before updating dependencies, check your application's calls against the llama.cpp 1.0.0 public headers and record every removed function that needs a replacement.

Source: https://gihyo.jp/article/2026/11/llama.cpp-v1.0.0

Contents