llama.cpp: Releases and Overview
LLM inference engine written in C/C++. Runs GGUF models on CPU and GPU (CUDA / Metal / Vulkan / ROCm) and underpins much of the local-LLM ecosystem, including Ollama, LM Studio and KoboldCpp.
Category: Inference engines and runtimes. Part of our engine and tool tracker.
At a Glance
| Item | Value |
|---|---|
| Repository | ggml-org/llama.cpp |
| License | MIT |
| Main language | C++ |
| GitHub stars | 128,500 (as of 2026-09-17) |
| Latest release | v0.8.0 (2026-10-28) |
| Install / docs | official documentation |
The project describes itself as: “LLM inference in C/C++”
Release History
Compiled by Local Model Watch from the project’s GitHub releases. Pre-releases are omitted. Where we wrote an article about a release, it is linked in the last column; smaller releases are tracked here only.
| Version | Released | Release notes | Our article |
|---|---|---|---|
| v0.8.0 | 2026-10-28 | GitHub | — |
| v0.7.0 | 2026-10-16 | GitHub | — |
| v0.6.0 | 2026-10-06 | GitHub | llama.cpp v0.6.0 Released with llama_batch_ext and Metal Boosts |
| v0.5.0 | 2026-09-24 | GitHub | llama.cpp v0.5.0 Released with Backend and Server Upgrades |
| v0.4.1 | 2026-09-15 | GitHub | llama.cpp v0.4.1 Released with Breaking Changes |
| v0.4.0 | 2026-09-05 | GitHub | — |
| v0.3.0 | 2026-08-25 | GitHub | — |
Speed by Version on Our CPU
Measured by Local Model Watch on our CPU-only server (Neoverse-N1, 3 threads) with llama-bench (512-token prompt, 128-token generation), always on the same model: Qwen3-0.6B-Q4_K_M.gguf from unsloth/Qwen3-0.6B-GGUF (revision 50968a4468). Versions like v0.5.0 ship no binaries, so we used the first official ubuntu-arm64 build that contains the version; “+N” is how many commits that build is ahead of it. These are CPU figures, not a guide to GPU speed, and small differences can be run-to-run noise (± is the standard deviation).
→ Scroll horizontally to see all columns
| Version | Released | Build measured | Prompt | Generation | Measured |
|---|---|---|---|---|---|
| v0.8.0 | 2026-10-28 | b11960 | 127.9 ± 0.1 tok/s | 42.4 ± 0.4 tok/s | 2026-10-29 |
| v0.7.0 | 2026-10-16 | b11669 | 128.4 ± 0.1 tok/s | 43.1 ± 0.2 tok/s | 2026-10-17 |
| v0.6.0 | 2026-10-06 | b11429 | 128.4 ± 0.6 tok/s | 43.0 ± 1.4 tok/s | 2026-10-07 |
| v0.5.0 | 2026-09-24 | b11146 | 129.9 ± 0.4 tok/s | 43.4 ± 0.5 tok/s | 2026-10-03 |
| v0.4.1 | 2026-09-15 | b10964 | 128.5 ± 1.1 tok/s | 43.4 ± 0.3 tok/s | 2026-10-03 |
| v0.4.0 | 2026-09-05 | b10809 | 129.9 ± 0.2 tok/s | 42.4 ± 0.7 tok/s | 2026-10-02 |
| v0.3.0 | 2026-08-25 | b10621 | 127.5 ± 0.9 tok/s | 43.2 ± 0.1 tok/s | 2026-10-02 |
Articles on Local Model Watch
- 2026-10-06 llama.cpp v0.6.0 Released with llama_batch_ext and Metal Boosts
- 2026-09-24 llama.cpp v0.5.0 Released with Backend and Server Upgrades
- 2026-09-15 llama.cpp v0.4.1 Released with Breaking Changes
Last updated 2026-10-29 (JST). Facts above come from the GitHub API; the one-paragraph summary was written by the site’s operator with the help of AI.