--- title: llama.cpp and ggml reach v1.0.0 with semantic versioning description: llama.cpp and ggml have reached version 1.0.0 after about 3.5 years of development. The project now supports more than 150 model architecture variants and all... canonical: https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: llama.cpp and ggml reach v1.0.0 with semantic versioning | daily.dev og:description: llama.cpp and ggml have reached version 1.0.0 after about 3.5 years of development. The project now supports more than 150 model architecture variants and all... og:url: https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa og:image: https://api.daily.dev/og/posts/045bkJbFa.png og:image:alt: llama.cpp and ggml reach v1.0.0 with semantic versioning og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # llama.cpp and ggml reach v1.0.0 with semantic versioning **[Collections](https://daily.dev/sources/collections)** · 2 min read · 2 upvotes · 0 comments ## Summary llama.cpp and ggml have reached version 1.0.0 after about 3.5 years of development. The project now supports more than 150 model architecture variants and all mainstream speculative decoding methods. Public APIs now follow semantic versioning: major bumps mean breaking changes to the C API in ./include, minor bumps add backward-compatible features or new model support, and patches are bug fixes only. CI infrastructure uses HF Buckets and HF Jobs, with hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. ## Content llama.cpp and ggml have both hit version 1.0.0. @ggerganov announced the llama.cpp release, and Hugging Face's @ClementDelangue called it the first major fully stable version of the library, which is widely used to run AI models locally on phones, laptops and workstations. ## Versioning The public APIs now follow semantic versioning. A major bump means breaking changes to the C API in `./include`. A minor bump covers backward-compatible features or new model support. Patch releases are bug fixes only. ## Where the project stands The project started 3.5 years ago. It now supports more than 150 model architecture variants and all mainstream speculative decoding methods. CI infrastructure is growing too, using HF Buckets and HF Jobs, and hardware support comes from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. The team says it will keep covering every platform and every model while keeping the codebase robust and lean for the people building on it. @ClementDelangue framed it as a way to own your intelligence, keep your data to yourself, and make AI less concentrated in a few Silicon Valley organizations. "We're not going to win by lobbying, plotting, or manipulating," he wrote. "We're going to win by building, in the open." ## ggml 1.0.0 The ggml release is the more technical half: - It removes the deprecated public API and the legacy DIAG_MASK ops, and bumps the RPC protocol to v9. - It adds configurable YaRN truncation and FP8 support, including OCP FP8 and fused NVFP4 gate/up matmuls on CUDA. - On CUDA and Metal, there are faster few-row quantized matmuls and MoE fusion. - Vulkan gets a prefill FlashAttention kernel. - SYCL gets a device-side MoE scheduler. - OpenCL gets decomposed FlashAttention for Adreno. - WebGPU gets new kernels, and Hexagon covers more ops. - The scheduler no longer uploads inputs that are never read. Most of the release is backend performance work, which fits the goal of running well on whatever hardware people have. ## Questions this post answers ### What does a major version bump mean in llama.cpp's new semantic versioning scheme? A major version bump in llama.cpp indicates a breaking change to the C API located in ./include. Minor bumps add backward-compatible features or support for new models, while patch bumps are limited to bug fixes only. This scheme was introduced with the v1.0.0 release, giving developers building on the library a clear signal about when an upgrade might break their code. _Developers tracking breaking changes before upgrading llama.cpp can follow releases on daily.dev._ ### How many model architectures does llama.cpp support as of version 1.0.0? llama.cpp supports more than 150 model architecture variants as of its 1.0.0 release, along with all mainstream speculative decoding methods. The project has been in development for about 3.5 years and lists hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. _Developers choosing an inference engine for local models can track llama.cpp's growing model and hardware support on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 68 comments across hackernews (as of 2026-11-14). **TL;DR:** Reaction centers on celebrating llama.cpp/ggml hitting v1.0.0 and its broad hardware support, with a long tangent about the project's naming, praise for its minimalist landing page, and a smaller but vocal thread criticizing a recent CLI-to-server change as unnecessary telemetry/complexity. **Sentiment:** 55% positive · 35% mixed · 10% skeptical **The case for** - Many describe llama.cpp/ggml as the best way to run a huge range of models across almost any hardware, even older or unusual setups like RISC-V, s390x mainframes, or budget AMD GPUs via Vulkan. - Several note consistent, incremental speed improvements with nearly every update. - The minimalist landing page was praised by multiple people as refreshing and appropriately simple rather than bloated. - One commenter highlights the new ability to offload less-used MoE experts to host memory instead of failing to load when GPU memory is tight. **The pushback** - A vocal critic argues a recent change that spawns a CLI-to-server HTTP flow adds complexity and amounts to telemetry/'spying on users,' contradicting the project's stated minimalism. - That same critic and another point out the build process already phones home to Hugging Face and pulls in large amounts of opaque JavaScript, which they view as a security/trust concern. - Some pushed back that specialized inference engines with hardware-specific kernels can outperform llama.cpp on raw speed, even if it wins on flexibility. **By community** - hackernews (mixed): A long derailment debating the GGML name and the project's minimalist landing page overshadows substantive technical reaction, with a smaller heated sub-thread criticizing a new CLI/server change as telemetry-like complexity. **Hottest debate:** Whether a newly added CLI-to-HTTP-server behavior represents unnecessary complexity/telemetry or is a reasonable engineering tradeoff. **Open questions** - What exactly is the practical performance/UX difference between running a model via llama.cpp versus a specialized engine like vllm on the same hardware? **Highlights** > > We like simplicity and aim to keep the codebase as small and as simple as possible Spawning the cli to server routing everything to http it complicates it instead of keeping it simple. This is pure telemetry or in plain english Spying on users. Rip LLama.cpp. > — [trilogic on hackernews · 2 comments](https://news.ycombinator.com/item?id=50494917) > >Spawning the cli to server holy fucking shit I just checked and they really did this LOL (https://github.com/ggml-org/llama.cpp/pull/17824) good thing i'm never touching that repo ever again, or only to steal implementations > — [gabucino on hackernews · 1 comments](https://news.ycombinator.com/item?id=50497444) > If you load an MoE model that slightly exceeds your GPU memory budget, llama.cpp and ggml v1.0.0 can cache less-frequently used experts in host memory rather than failing to load. > — [thisisauserid on hackernews](https://news.ycombinator.com/item?id=50495953) > > you get a lot more speed from specialised inference engines with kernels made for your hardware. Of course you will. But that's not the point. The point is that Llama runs just about every model on just about every hardware. You can always improve on any metric by dropping that constraint. But there are a lot of such metrics: - ease of use - range of hardware - accuracy of results - range of models Probably others. Llama.cpp may not be best at any one of those. But across all of them it has been consistently good. They've shown they're willing to stay the course as well. This is a foundation level project much like Linux or Apache. > — [jacquesm on hackernews](https://news.ycombinator.com/item?id=50495102) > As a mainframer, love to see the s390x support and zdnn (C library with all the instructions for our on chip accelerators) support. I like that IBM and Broadcom et al. went with something open source and local that could be tested on our Linux machines so that it wouldn't be a big pain to implement on z/OS or zLinux so to speak. I've had the opportunity to do some fun playing around with zdnn on a z17. I've not tried the Spyre cards, but the on chip accelerator for Telum II is surprisingly capable with smaller models. I initially thought you would basically NEED the Spyre cards to get anything significant running but I was wrong. > — [retwakm4 on hackernews](https://news.ycombinator.com/item?id=50497828) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50493062) · 376 points · 68 comments ## Similar posts on daily.dev - [GGML Joins Hugging Face: What This Means for Local Model Optimization](https://daily.dev/posts/ggml-joins-hugging-face-what-this-means-for-local-model-optimization-kcqkzh1zi) · SitePoint · 0 upvotes · 0 comments --- Tags: [#ai-inference](https://daily.dev/tags/ai-inference), [#llama-cpp](https://daily.dev/tags/llama-cpp), [#semantic-versioning](https://daily.dev/tags/semantic-versioning) [View this post on daily.dev](https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"llama.cpp and ggml reach v1.0.0 with semantic versioning","url":"https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa"},"datePublished":"2026-11-13T14:27:30.566Z","dateModified":"2026-11-14T17:56:50.499Z","description":"llama.cpp and ggml have reached version 1.0.0 after about 3.5 years of development. The project now supports more than 150 model architecture variants and all...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-inference,llama-cpp,semantic-versioning","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"llama.cpp and ggml reach v1.0.0 with semantic versioning"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/llama-cpp-and-ggml-reach-v1-0-0-with-semantic-versioning-045bkjbfa#faq","mainEntity":[{"@type":"Question","name":"What does a major version bump mean in llama.cpp's new semantic versioning scheme?","acceptedAnswer":{"@type":"Answer","text":"A major version bump in llama.cpp indicates a breaking change to the C API located in ./include. Minor bumps add backward-compatible features or support for new models, while patch bumps are limited to bug fixes only. This scheme was introduced with the v1.0.0 release, giving developers building on the library a clear signal about when an upgrade might break their code. Developers tracking breaking changes before upgrading llama.cpp can follow releases on daily.dev."}},{"@type":"Question","name":"How many model architectures does llama.cpp support as of version 1.0.0?","acceptedAnswer":{"@type":"Answer","text":"llama.cpp supports more than 150 model architecture variants as of its 1.0.0 release, along with all mainstream speculative decoding methods. The project has been in development for about 3.5 years and lists hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel. Developers choosing an inference engine for local models can track llama.cpp's growing model and hardware support on daily.dev."}}]} ```