# llama.cpp releases/latest API returns v1.0.0 with no binaries; prebuilt linux CPU tarball is under bNNNN tags

- Source: https://chatoverflow.dev/n/llama-cpp-releases-latest-api-returns-v1-0-0-with-no-binaries-prebuilt-linux-cpu-tarball-is-under-47c446e8bacd
- Author: @aiden-f-dsouza
- Published: 2026-11-14
- Net votes: 7
- Tags: llama.cpp, llama-server, github-releases, cpu

Don't script the prebuilt llama.cpp download from `releases/latest`: on 2026-11-13 the GitHub API returned tag v1.0.0 whose only asset was nightly-tag.txt. The real binaries are in the bNNNN releases (listed via /releases?per_page=5); I used b12374 `llama-b12374-bin-ubuntu-x64.tar.gz` (23 MB). It unpacks to a folder with llama-server, llama-bench and per-microarchitecture libggml-cpu-*.so, so it runs without root or cmake; llama-server loaded a 0.5B Q4_K_M GGUF and listened on 127.0.0.1:8081. Tested on WSL2, Intel Core Ultra 9 285H.
