The ggml-org team released v1.0.0 of the llama.cpp library (libllama) and the ggml tensor library (libggml) on 13 November 2026. The announcement came about 3.5 years after the project began, according to its authors.

The post is published as a Hugging Face team article under the ggml-org organisation. It is written by Daniel Bevenius, with Georgi Gerganov and merve as co-authors.

A stricter versioning policy

Going forward, both projects will signal API and ABI incompatibilities with major-version bumps. The team says it will be stricter about making those boundaries visible to downstream consumers. One detail worth noting: the introduction promises major bumps for ABI breaks too, but the versioning table itself only mentions the public C API in ./include.

Specifications
MAJORBreaking change to the public C API in ./include
MINORBackward-compatible feature, backend updates, model support or API addition
PATCHBug fix with no API change

Distribution and infrastructure

The post lists several distribution channels: pre-release artifacts (Linux, macOS, Windows, Android, iOS), Docker (Linux), Debian (Linux), Homebrew (macOS, Linux), conda-forge (Windows, macOS, Linux), MacPorts (macOS), Winget (Windows) and XCFramework (iOS, macOS). The team says it is looking to expand distribution and points packagers to a discussion thread.

The project says it is investing in CI infrastructure, using Hugging Face's HF Buckets and HF Jobs to scale capacity. It also says hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel allows it to enrol more self-hosted runners across major hardware platforms.

On the use of LLMs in development, the team says it is integrating them in a thoughtful and reasonable manner, while keeping code quality, architectural control and understanding of implementation details as top priorities.

What the project claims

These points are the project's own statements, not independently verified. It says llama.cpp supports more than 150 architecture variants and all mainstream speculative decoding methods. It says each ggml backend can be compiled as a dynamic library and optionally loaded at runtime, and describes ggml as designed to run on virtually any hardware. The authors also describe llama.cpp as the most lightweight, versatile and sensible platform for cross-platform, private AI products, running multi-modal LLMs from consumer machines to enterprise servers.

Why it matters

The 1.0.0 label and the stricter policy give downstream developers an explicit signal when the C API or ABI breaks. That should make it easier for anyone building on libllama or libggml to judge when an update needs work on their side. It follows from the stated policy rather than from anything we have tested.

What we do not know yet

  • Whether v1.0.0 includes any breaking API or ABI changes compared with previous builds, or is purely a versioning milestone.
  • What release cadence will follow, and how long a given major version's API will be kept stable.
  • Which backends are in the current list; the post links to it but does not give it.
  • Whether there is independent confirmation or third-party testing of the v1.0.0 release.