ggml-org releases ggml and llama.cpp v1.0.0 with stable APIs
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
ggml-org released ggml and llama.cpp v1.0.0 on Nov 13, pruning deprecated APIs and adopting strict semantic versioning.
The stable C API affects the many apps and packagers built on these libraries.
Background
ggml and llama.cpp are open-source projects from ggml-org, the team behind the llama.cpp local-AI inference software; ggml is the underlying library, described by its maintainers as designed to run on virtually any hardware in existence. Georgi Gerganov, the creator of llama.cpp, announced in early November that a first major v1.0.0 release was coming within weeks. The 1.0 milestone arrived 3.5 years after the project's initial commit, and Gerganov's company ggml.ai and its team joined Hugging Face in February 2026, with the projects staying open source. The team said the project has grown a lot and that many downstream applications already depend on its work, making a flexible but stable API a priority.
How it unfolded
- Nov 4, 2026Georgi Gerganov announced llama.cpp v0.9.0 and said the project was gearing towards the first major v1.0.0 release in the next couple of weeks.
- Nov 13, 2026ggml-org released ggml v1.0.0 (libggml) and llama.cpp v1.0.0 (libllama); the ggml release was published at 08:34.
- Nov 13, 2026ggml 1.0.0 pruned the deprecated public API, removed the legacy DIAG_MASK_INF/DIAG_MASK_ZERO ops, added configurable YaRN truncation and FP8 support, and bumped the RPC protocol to v9 to reject incompatible peers.
- Nov 13, 2026llama.cpp 1.0.0 removed 35 deprecated functions and the warmup and defragmentation-threshold settings from llama.h, promoted the NextN speculative-decoding API into the main header, and renamed llama_set_embd_state_src to llama_set_ctx_state_src.
- Nov 13, 2026llama.cpp 1.0.0 added OCP FP8 key-value caches, a host-memory cache for mixture-of-experts weights with a configurable CPU share, three new text architectures including Kimi-K3.1-Pro, a /v1/audio/speech endpoint, and hardening of llama-server against SSRF, CORS and DNS-rebinding attacks.
- Nov 13, 2026Both projects adopted strict semantic versioning: major bump for breaking public C API changes, minor for backward-compatible features and model support, patch for bug fixes.
- Nov 13, 2026The release shipped across Linux, macOS, Windows, Android and iOS, with builds through Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and an XCFramework for Apple platforms.
- Nov 14, 2026CASRAI reported the 1.0.0 releases and their compatibility boundaries for research software maintainers.
Key terms
- semantic versioning
- A version-numbering scheme using a three-part Major.Minor.Patch number, with optional prerelease and build metadata tags.
- SSRF
- A security flaw letting an attacker make a vulnerable server send requests to internal or external systems.
Source: Wikipedia
Who’s saying what
- Official
- Creator Georgi Gerganov said every new integration means more developers depending on the library interfaces, so a flexible but stable API is paramount.
- Official
- The ggml-org team said the project has grown a lot in 3.5 years and that many downstream applications already depend on its work.
- Analysts
- CASRAI framed the release around its compatibility boundaries for research software maintainers.