- Sources: ggml discussion 1579, llama.cpp v0.1.0 release, llama.cpp releases API, HN discussion
- Summary: llama.cpp tagged v0.1.0 on 2026-08-17 and published v0.1.2 at 10:23:10 UTC on 2026-08-18. The v0.1.0 release page renders navigation chrome and an empty release body, so the release fields were read from the releases API instead. v0.1.2 is marked prerelease, the b-number nightly b10488 is still the latest non-prerelease build, and ggml discussion 1579, opened by ggerganov on 2026-08-07, states that semantic versioning for llama.cpp is not yet implemented, which is the ground for the developing status. The bump rules and cadence proposed in that thread are maintainer discussion rather than settled policy.
- Why it matters: Distributions and package managers currently pin a 125,000-star inference runtime to nightly commit-number builds, and a Debian maintainer in the thread states that a version scheme guaranteeing the vendored ggml copy matches a released ggml is what would let llama.cpp ship as a library with a SOVER in public search paths.
- Follow-up: Track whether semantic versioning becomes the non-prerelease channel and whether the b-number nightlies are retired.
send feedback on this story