llama.cpp finally ships a v0.1.0
Three years after first commit, the project that made local LLMs possible ships a stable release.
llama.cpp shipped v0.1.0 this week. Not a typo. The project that runs Llama 3.3 on a Raspberry Pi, that spawned a thousand local inference tools, that forced Apple to take ARM ML seriously, finally cut a semantic version. The first commit was February 2023. Thousands of forks, millions of downloads, every Mac with 16GB RAM suddenly running frontier models in terminal. The version number stayed at zero the whole time. This is how infrastructure works now. Ship fast, break nothing, version numbers are a formality. The real signal is GitHub stars and daily active forks. llama.cpp crossed that threshold years ago. The 0.x label was technical honesty, not hesitation. The changelog is polite and brief. Quantisation improvements, better metal support, CPU dispatch fixes. Nothing groundbreaking because the groundbreaking stuff already shipped. This release is housekeeping. What changed is project confidence. Someone finally decided the API surface is stable enough to promise compatibility. That matters more than any single feature. If you are embedding llama.cpp into a product, you can now pin a version and sleep. The timing is curious. OpenAI ships o3 in the same week, Meta releases Llama 4 benchmarks. The big labs are sprinting. Meanwhile the scrappy C++ project that democratised inference is cleaning up the repository structure. Both matter. One is just quieter.
Source: Llama.cpp v0.1.0