Skip to content

Reference version

The course uses v0.4.1; llama.cpp is at v0.7.0

As of 10/28/2026 the latest stable release is v0.7.0 (10/15/2026) and the latest build is b11936. The change that affects the course commands: since v0.7.0 llama serve listens on port 9931 instead of 8080 (the Docker images stay on 8080 inside the container). Replace 8080 with 9931, or add --port 8080 when you start the server. New features introduced after v0.4.1: --moe-cache-mib (GPU cache for MoE experts kept in RAM), LiliCorr speculative decoding, ggml 0.27; for these see the release notes.

llama.cpp publishes several builds a day. To avoid chasing them, the course uses one precise version:

Field Value
Build used in the course b11046: llama version prints version: 0.4.1-dev (build 11046, commit 60081bb2b)
How to get it official installer curl -LsSf https://llama.app/install.sh \| sh (it is the build the installer shipped on 09/22/2026); tarball llama-b11046-bin-*; sources: git checkout b11046
Closest stable release v0.4.1 of 09/14/2026 (corresponds to build b10964)
ggml 0.24.0
Latest build seen when the course was written b11095 (09/22/2026)

If you use a different version

That is fine: almost everything works the same. The things that change most often are:

  • binary names: since 2026 there is the single llama command with subcommands (llama cli, llama serve, llama bench…). The old names llama-cli, llama-server, llama-bench still exist in the tarballs and in Docker, and many online tutorials use them: they are equivalent;
  • removed or renamed flags: in 0.4.1 --mmap, --mlock, --direct-io became --load-mode; always check llama <comando> --help;
  • server port: up to 0.6 llama serve uses 8080, from 0.7.0 it uses 9931; with --port you choose;
  • supported models: every build adds architectures; a new GGUF may require a new build.
flowchart TB
    M[master: continuous commits] --> B["build bNNNNN<br/>(several a day, binaries for all platforms)"]
    B -->|every ~2 weeks| V["release vX.Y.Z<br/>(stable, with release notes)"]
    B -->|llama.app installer| I[~/.local/bin/llama]
    V -->|winget, brew, conda| P[package managers]

How to update the course to a new version

  1. Install the new build (llama update), write down llama version.
  2. Rerun labs 01, 02, 04 and 05 (they are the ones with the most commands).
  3. Update the table above and the update dates of the pages you touched.