llama.cpp and ggml reached 1.0, and breaking changes to their C API will now require a major version bump
The llama.cpp team released version 1.0.0 of llama.cpp and ggml, with a public commitment to semantic versioning.
What happened
In a post on the Hugging Face blog dated November 13, the ggml-org team announced v1.0.0 of libllama and libggml. Going forward, both projects will signal API and ABI incompatibilities with major version bumps. The rules are spelled out: a breaking change to the public C API in the include directory is a major bump, a backward compatible feature, backend update, model support or API addition is a minor bump, and a bug fix with no API change is a patch.
Latent Space's AINews adds that the 1.0 release removed 35 deprecated functions from llama.cpp's public header, and that users running mixture-of-experts models larger than their GPU memory can now reserve a cache for frequently used experts.
The post lists distribution through Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and an XCFramework for Apple platforms. It says llama.cpp supports more than 150 architecture variants and all mainstream speculative decoding methods, and that the team is expanding CI capacity with hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. Hugging Face also announced Llama desktop apps for macOS and Windows built on top of llama.cpp.
My take
For anyone running models locally for a client, this is mainly an operations change. Until now, pulling a newer build of llama.cpp could break a wrapper or a server integration without much warning. With semantic versioning, you can pin to 1.x, take minor and patch updates with less worry, and treat a 2.0 as a planned migration.
This matters most for the cases where I'd recommend local inference in the first place: data that can't leave the client's network, steady high volume classification or extraction where per-token API costs add up, or offline sites. In those setups the model server is a dependency like any other, and it should be versioned, tested and upgraded on a schedule.
My practical advice is to pin the exact version in your Docker image or package config today, since the release removed deprecated functions. If your integration calls the C API directly, check your code against the 1.0 header before upgrading.
More posts
- Open OCR models now read 1,000 pages for under $0.10, and better serving saved more than switching modelsNov 20, 2026
- Gemini 4 Neon costs $3 in and $15 out per million tokens and tops Zapier's AutomationBench at 56.3%, though it uses more output tokens per taskNov 20, 2026
- Gemini API's Tool Retrieval loads tool schemas only when the model asks for them, so agents with hundreds of tools send fewer input tokensNov 20, 2026
