llama.cpp and ggml reach version 1.0.0, pruning deprecated APIs

llama.cpp, the open source engine that many desktop and mobile apps use to run language models locally, has reached version 1.0.0. The v1.0.0 release on GitHub was published on 13 November alongside a matching ggml v1.0.0 release for the tensor library underneath it.

What changes for developers

The llama.cpp release notes say v1.0.0 removes the deprecated public API and promotes the NextN, or speculative decoding, API into the main llama.h header. It adds OCP FP8 KV caches, a host-memory cache for mixture-of-experts weights with a configurable CPU share, and loading of external PLE shards.

On the server side, the notes list fixes for CORS origin handling, DNS rebinding and a CORS proxy SSRF issue, plus an OpenAI-compatible /v1/audio/speech endpoint. The bundled web UI has been redesigned with chat groups and a markdown editor.

The ggml 1.0.0 notes say the library prunes its deprecated API and the legacy DIAG_MASK operations, bumps its RPC protocol so incompatible peers are rejected, and adds configurable YaRN truncation and FP8 support, with performance work across CUDA, Metal, Vulkan, SYCL, OpenCL, WebGPU and Hexagon backends.

Why the version number matters

The AI Compass digest reports that from 1.0.0 both projects follow strict semantic versioning for their public C headers, so breaking API or ABI changes will require a major version bump. We could not load the project’s Hugging Face blog post about the release, so that versioning detail comes from AI Compass’s summary rather than a page we read directly.

Analysis: for apps that wrap libllama, a stable header contract is the headline. Removing the deprecated API means some older integrations will need code changes to move to 1.0.0.

This report is not legal, financial, or health advice. See the disclaimer.